The Court of Justice decidedly jumps on the procurement protectionism bandwagon, creating legal uncertainty along the way (C‑652/22)

** This was first published as an Op-Ed for EU Law Live on 25 Oct 2024 (formatted pdf version). I am reposting it here in case of broader interest. **

By falling just short of mandating a complete ban on access to EU procurement by third-country economic operators not covered by international agreements, in Kolin Inşaat Turizm Sanayi ve Ticaret (C-652/22) (‘Kolin’), the Court of Justice has suddenly crystallised a change in EU procurement-related trade policy. Kolin will have many, and potentially quite problematic, practical ramifications. In this Op-Ed, I reflect on the broader context in which the Court has reached its position. I also stress why I think this is an extraordinary case of judicial activism, as the Court of Justice has gone out of its way to answer a question it had not been asked. I show how the position taken by the Court goes well beyond the settled (expectations of) legal interpretation under the EU procurement rules, creates significant legal uncertainty, and brings into question the future of recently developed EU procurement-related trade law instruments. I had previously criticised the Opinion of AG Collins in this case, and that earlier comment contains additional problems in the position taken by the Court.

Some context

The EU historically kept its procurement markets open to international trade, not only through the inclusion of procurement commitments in free trade agreements (FTAs) and championing developments under the World Trade Organisation Government Procurement Agreement (GPA), but also by being one of the most open procurement markets in the world. The traditional policy approach was largely one of economic persuasion; by granting or tolerating access to its internal market for public contracts, the EU expected to obtain (legal or de facto) reciprocity. Traditionally, EU procurement law has thus not prevented access from third-country economic operators. While it has stressed the bindingness of commitments towards economic operators covered by the FTAs or GPA (‘agreement-covered economic operators’), it has also tolerated participation of other ‘third-country economic operators’.

This is clear in the 2014 public procurement package, and especially in the context of procurement in the water, energy, transport and postal services sectors, where the Utilities Directive (2014/25/EU) explicitly contains, on the one hand, a reminder of the obligation of ‘no less favourable treatment’ of agreement-covered economic operators (Art. 43) and, on the other, rules on the treatment of tenders including products originating in third countries (Art. 85), and on the broader procurement relations with those third countries (Art. 86). This reflects the traditional position of pragmatic tolerance of access by third-country economic operators in search for (legal or de facto) reciprocity.

This policy started to shift in parallel with the negotiations of the 2014 public procurement package, and eventually resulted in the adoption, in 2022, of the International Procurement Instrument (IPI) and the Foreign Subsidies Regulation (FSR). In simple terms, these instruments grant powers to the Commission to restrict or ban access to procurement markets by third-country economic operators and, in the case of the FSR, this includes agreement-covered operators. These instruments signal a significant ‘hardening’ of the EU’s stance towards trade-related procurement and, in my view and the view of others, a shift towards economic protectionism.

It is important to stress that the IPI and FSR are built on the premiss that EU procurement law does not mandate the absolute exclusion of third-country economic operators. This was already clear in the 2019 Guidance on the participation of third-country bidders and goods in the EU procurement market, which stressed that third-country economic operators ‘do not have secured access to procurement procedures in the EU and may be excluded’ (my emphasis). The IPI was precisely developed to create a process for such exclusion. The summary of the IPI stresses that it gives the Commission the power to ‘impose, as a last resort, IPI measures to restrict access to EU public procurement procedures for businesses, goods and services from the non-EU countries concerned’ (my emphasis). The non-EU countries concerned being only those with which the EU does not have a procurement agreement, as the IPI does not apply to agreement-covered economic operators (Art. 1(3)). Similarly, the FSR grants the Commission the power to prohibit the award of a contract to an economic operator that has benefitted from a foreign subsidy distorting the internal market, in this case without distinguishing between agreement-covered and other third-country economic operators (Art. 31(2) and recital 45).

It goes without saying, but seems to need spelling out, that if the EU procurement rules—and the Utilities Directive in particular—had already banned access from third-country economic operators, the IPI would be entirely redundant and the FSR would only be relevant in relation to agreement-covered economic operators. This does not seem to be the view of the Court, though, as we will see that the Judgment in Kolin markedly deviates from the—until now majoritarian, if not universal—view that EU procurement law tolerated participation by third-country economic operators.

Getting out of its way to answer a question that was not asked

Before getting into the substance of the Judgment, it is worth highlighting third-country economic operators’ access to EU procurement markets was by no means an issue the Court necessarily had to rule on in Kolin. Indeed, for me, one of the striking aspects of Kolin is that the Court has gone out of its way to answer a question it was not asked.

The case concerned a rather mundane set of circumstances where the contracting authority allowed its preferred tenderer to provide additional documentation demonstrating that it met the relevant technical and professional capacity requirements, after the original award decision had been quashed for shortcomings in the initial documentation. The preliminary reference thus concerned a set of questions seeking to ascertain the limits on the contracting authority’s discretion to request and consider such supplementary documentation at such late stage of the procedure, in particular in relation to a potential breach of the principles of equal treatment and non-discrimination—if necessary, interpreted in relation to the right to good administration under Art. 41 of the Charter (although this does not seem to have been raised in the case).

The focus of the dispute could thus squarely be placed on the behaviour of the contracting authority and its duties under the Utilities Directive. From the perspective of the adequate interpretation and application of EU (procurement) law, it is irrelevant how the case reaches the Court of Justice, as the core legal issue concerns the constraints of the behaviour of a contracting authority that has positive duties under the Utilities Directive. There is no question that the Utilities Directive applied to the conduct of the specific tender procedure. From this perspective, there is limited room to question the applicability of EU (procurement) law and the relevance of the questions posed by the referring court. The case could have been decided like others in a long-running saga of case law on clarifications of tenders.

However, the Court took a different approach and decided to focus on the fact that the complainant in the domestic dispute is a third-country economic operator (from Türkiye) (Kolin, para. 39). This shifted the focus from the duties constraining the contracting authority under the Utilities Directive to the rights of the complainant. In the end, the question that Kolin addresses is whether third-country operators can participate in tenders for public contracts covered by EU law and, if so, under which conditions. This question had not been asked by the referring court (Kolin, para. 40). This marked shift in approach not only opened a Pandora’s box of issues concerning the interpretation of the Utilities Directive and its fit with other instruments of procurement-related trade policy, but also provided the (self-generated) opportunity for the Court to decidedly jump on the protectionism bandwagon.

Given how purposefully the Court grabbed the opportunity to answer a question it had not been asked, the way in which it addressed the question becomes particularly problematic because, in falling short of explicitly and unqualifiedly mandating exclusion of third-country operators, the Court has created legal uncertainty with potentially significant practical and conceptual ramifications.

Participation without enforceable rights?

The Court of Justice has taken the view that, regarding third-country economic operators, ‘although EU law does not preclude those economic operators, in the absence of exclusion measures adopted by the European Union, from being allowed to participate in a public procurement procedure governed by Directive 2014/25, it does, however, preclude those economic operators from being able, in the context of their participation in such a procedure, to rely on that directive and thus to require that their tender be treated equally to those submitted by tenderers from Member States and by the tenderers from third countries referred to in Article 43 of that directive’ (Kolin, para. 45, emphasis added). The justification given the Court is that allowing third-country operators to rely on the provisions of the Directive to make claims of equal treatment ‘would have the effect of conferring on them a right to no less favourable treatment, contrary to Article 43 of that directive, which limits the benefit of that right to economic operators from third countries which have concluded with the European Union an international agreement’ (Kolin, para. 46, emphasis added). The Court further states that

… To interpret that provision differently and thus to render unlimited the personal scope of that directive would be tantamount to guaranteeing economic operators of those third countries equal access to public procurement procedures in the European Union. For the reason set out in paragraph 46 of the present judgment and as is also stated, now, in recital 10 of the IPI Regulation, Directive 2014/25 must be understood as meaning that the access of economic operators of those third countries to public procurement procedures in the European Union is not guaranteed and that those operators may be excluded from them (Kolin, para. 47, emphasis added).

Ultimately, then, the Court established that ‘in a situation … involving the participation, as accepted by the contracting entity, of a Turkish economic operator in a public procurement procedure governed by Directive 2014/25, that operator cannot rely on … that directive in order to challenge the decision awarding the contract concerned’ (Kolin, para. 51).

The Court seems to be aiming for an impossible equilibrium between the traditional position of de facto tolerance and the preservation of the rights of agreement-covered economic operators. In my view, in falling short of mandating a ban on participation and opting instead for the position that EU procurement law foresees the possibility for third-country economic operators to participate with no enforceable rights, the Court creates two problems.

The first problem arises from the recognition by the Court that, ultimately, treating the tenders submitted by third-country economic operators equally to those submitted by tenderers from Member States and by the tenderers from agreement-covered economic operators limits the benefit of the rights of the latter. This raises the question whether, functionally, the mere tolerance of participation by third-country economic operators on an equal basis would also be an infringement of Art. 43 of the Utilities Directive because contracting authorities would not be preserving the benefits expected to the rights of agreement-covered economic operators (and tenderers from Member States?). Is the logical implication of the position stated by the Court that Art. 43 of the Utilities Directive requires ‘less favourable’ treatment to be granted to tenders from third-country economic operators?

The second, and related, problem is that the Court limits its judgment to the impossibility for third-country economic operators to rely on EU procurement law in order to challenge the decision awarding the contract concerned. But does this extend to challenges based on national law or other grounds?

The Court goes on to ‘fix’ those two problems. But, in my view, it does so in a manner that triggers more problems than it solves.

A problem actually pushed down to the domestic level, on a case-by-case basis

The issue of the treatment of third-country economic operators’ participation at the domestic level is also addressed by the Court of Justice in Kolin. After rehearsing the position under the Treaties and in case law, the Court reminds us that

…only the European Union has competence to adopt an act of general application concerning access, within the European Union, to public procurement procedures for economic operators of a third country which has not concluded an international agreement with the European Union guaranteeing equal and reciprocal access to public procurement, by establishing either a system of guaranteed access to those procedures for those economic operators or a system which excludes them or provides for an adjustment of the result arising from a comparison of their tenders with those submitted by other economic operators (Kolin, para. 61).

And that

… the European Union has not empowered the Member States to legislate or adopt legally binding acts concerning access to public procurement procedures for economic operators of a third country which has not concluded an international agreement with the European Union (Kolin, para. 62).

Logically, this should have extended the ban on relying on EU law to challenge procurement decisions to the national level, and the consistent position would have been that Member States cannot enable ‘no less favourable treatment’ claims under their domestic provisions.

However, in a surprising twist of argumentation identifying contracting authorities as something potentially separate from the Member State to which they belong, the Court stressed that

In the absence of acts adopted by the European Union, it is for the contracting entity to assess whether economic operators of a third country which has not concluded an international agreement with the European Union guaranteeing equal and reciprocal access to public procurement should be admitted to a public procurement procedure and, if it decides to admit them, whether provision should be made for an adjustment of the result arising from a comparison between the tenders submitted by those operators and those submitted by other operators (Kolin, para. 63, emphasis added).

Weirdly, this comes to empower each and every single contracting authority to take its own decisions on whether to tolerate participation of third-country economic operators. It can be that this will become a red herring, though, given the Court’s framing of the obligation to generate some degree of ‘less favourable’ treatment in case contracting authorities do decide to tolerate participation.

DIY ‘less favourable treatment’

In its final passages, the Court addresses the issue of the functional prohibition of no less favourable treatment for third-country economic operators. The Court indicates that

… it is open to the contracting entity to set out, in the procurement documents, arrangements for treatment intended to reflect the objective difference between the legal situation of those operators, on the one hand, and that of economic operators of the European Union and of third countries which have concluded such an agreement with the European Union, within the meaning of Article 43 of that directive, on the other hand.

In any event, national authorities cannot interpret the national provisions transposing Directive 2014/25 as also applying to economic operators of third countries which have not concluded such an agreement with the European Union, which have been admitted, by a contracting entity, to participate in a procedure for the award of a public contract in the Member State concerned, as otherwise the exclusive nature of the European Union’s competence in that area would be disregarded.

While it is conceivable that the arrangements for treatment of such operators should comply with certain requirements, such as transparency or proportionality, an action by one of those operators seeking to complain that the contracting entity has infringed such requirements can be examined only in the light of national law and not of EU law (Kolin, paras 64-66, emphases added).

This is, in my view, the most problematic part of the Kolin Judgment because it will be almost impossible to implement in any way that does not completely exclude participation by third-country economic operators. Assessing all implications of this passages exceeds the possibilities of this Op-Ed, as they concern, among other issues, how ‘less favourable’ must the treatment be and how does that square with approaches under the IPI and FSR, the extent to which EU and agreement-covered economic operators can challenge such treatment and claim that it does not adequately ‘reflect the objective difference between the legal situation’ of the different competitors, the interaction between the ECHR (in particular in relation to due process rights) and domestic rules that, by virtue of being excluded from the scope of the EU procurement directives may be seen not to be covered by the Charter (although I have my doubts that the Court could say, without departing from its prior case law) that the mandated ‘less favourable treatment’ of third-country economic operators is not within the scope of EU law for the purposes of the application of the Charter, the impact on domestic public and administrative law doctrines, the impact on current procurement procedures and procurement challenges, etc.

Final thoughts

Overall, I think Kolin is a bad Judgment. I think it does not make sense from a policy perspective because it crystallises an absolutist position that was not being pursued by the European Commission or the EU legislators. More importantly, in its own terms, I think it is a bad Judgment because the Court does not quite finish the job by establishing an absolute ban on participation. The position that domestic legislation is pre-empted by the EU’s exclusive competence but case-by-case decisions by contracting authorities are allowed as long as they ensure a degree of objective less favourable treatment that adequately reflects the objective difference between the legal situation of the different competitors (what a test to apply…) has opened a Pandora’s box that will have long lasting negative consequences.

Looking further ahead, it could well be that Kolin results in an explicit prohibition of third-country participation in the new procurement directives that are just about being developed. In that case, the EU legislator would have to take the explicit step the Court has tried to avoid or conceal. However, it could also be that it triggers a very complex approach to seeking to consolidate the ‘participation without rights and in less favourable conditions’ approach followed by the Court. Given the context of judicial activism in which Kolin has emerged, I keep wondering why did the Court not simply state that preserving the benefits arising from Art. 43 and the EU’s international commitments requires exclusion of third-country economic operators. I would still have thought it was a bad case on policy or normative grounds, but it would at least have made more legal sense and avoided creating a tidal wave of legal uncertainty, especially if the Court restricted the effects of the interpretation to the future.

A missed opportunity to provide meaningful clarification on state aid analysis of procurement compliance and some problematic ‘obiter dicta’ (C-28/23)

By Arne Müseler / www.arne-mueseler.com, CC BY-SA 3.0 de, https://commons.wikimedia.org/w/index.php?curid=149888646.

On 17 October 2024, the European Court of Justice (ECJ) delivered its preliminary ruling in NFŠ (C-28/23, EU:C:2024:893). The case was very interesting in three respects. First, in addressing some aspects of the definition of public works contracts that keep coming up in litigation in relation to relatively complex real estate transactions. Second, in addressing the effects of a State aid decision on the assessment of compliance with procurement law of the legal structure used to implement the aid package (including the treatment from a procurement perspective of put options as State aid measures). Third, in addressing some limits on the ‘strategic’ use of remedies by contracting authorities that have breached procurement law. Moreover, the case raised questions on the extent to which the parties to a dispute leading to a request for a preliminary reference can seek to clarify in front of the ECJ the underlying circumstances of the dispute, where the referring court has presented an incorrect or biased fact pattern.

The case indeed raised interesting issues and AG Campos Sánchez-Bordona delivered a promising Opinion that would have enabled the ECJ to provide helpful clarifications in those respects. However, in its NFŠ Judgment, the ECJ has not only missed that opportunity but also made some sweeping statements that could be problematic from the perspective of the interaction between State aid and procurement law.

I should from the outset disclose again that I was involved in the case. At the request of NFŠ, I wrote an expert statement addressing some of the issues before the ECJ. This may, of course, have affected my view of the case. However, I hope the comments below will help put the case in perspective and highlight the need to take some of the statements made by the Court with more than a pinch of salt. Actually, given the peculiar circumstances of the NFŠ case, I argue that they need to be considered as mere ‘obiter dicta’.

Background

I detailed the background of the case in my earlier comment on the AG Opinion, but it is helpful to restate the key issues here.

In 2013 the Slovak Government granted State aid to NFŠ to support the construction of the national football stadium in Bratislava. However, that State aid package was not considered sufficient and work did not start. The State aid measure was then revised in 2016 (the ‘grant agreement’), and the Slovak Government also granted NFŠ a unilateral put option to sell the stadium to the State, under certain conditions, during the five years following its completion (the ‘agreement to enter into a future sales agreement’ or ‘AFSA’).

Upon notification of the revised aid package, the Commission declared those measures to be compatible with the internal market by State aid Decision SA.46530. The State aid Decision made two important explicit points. First, it confirmed that the put option allowed NFŠ ‘to sell the Stadium back to the State in case it wishes to do so. Should the beneficiary decide to exercise the option, the Stadium would become a property of the State’ (para 22). The State aid Decision also explicitly stated that ‘The construction works financed through the grant … will be subject to a competitive process, respecting the applicable procurement rules’ (para 8).

Once the stadium was built, NFŠ exercised the put option. The Slovak Government decided not to purchase the stadium and it instead challenged the compatibility with EU law of the State aid package due to a fundamental breach of procurement law. The Slovak Government argued that the agreements were null and void because, combined and from the outset, the grant agreement and AFSA would have had the unavoidable effect of getting the stadium built and transferred to the State, and thus covered up the illegal direct award of a public works contract to NFŠ. This part of the dispute concerned the definition of ‘public works contracts’ under Directive 2014/24/EU (issue 1).

Relatedly, the Slovak Government stated that despite containing explicit references to the tendering of the construction of the stadium, the State aid Decision cannot preempt a fresh assessment of the compliance of this legal structure with EU procurement rules. Perhaps surprisingly, this position was supported by the European Commission in its submissions and at the hearing, where the Commission denied that the explicit mention of compliance with procurement law formed an integral part of its assessment of the compatibility of the set of agreements with EU internal market law. This was a crucial issue and the outcome of this case could have provided much needed clarity on the extent to which the Commission does, and indeed must, take procurement law into account in the assessment of State aid measures that involve the award of public contracts. This part of the dispute thus concerns the effect of State aid decisions relating to aid packages with a procurement element (issue 2).

Finally, the Slovak State sought confirmation of the possibility of having the ineffectiveness of the grant agreement and AFSA recognised ex tunc under domestic law, without this being a breach of the Remedies Directive. This relates to the ‘strategic’ use of procurement remedies by contracting authorities that have breached procurement law (issue 3).

In this post, I will focus on issues 1 and 2.

Framing: Directive 2004/18/EC, Directive 2014/24/EU, or it does not matter?

One preliminary issue worth highlighting is that the timeline of the case created the issue whether the 2004 or the 2014 procurement Directive applied. The initial grant agreement was signed in 2013, but the final grant agreement and AFSA were signed in 2016. On this point, despite taking opposite views (AG Campos focused on the 2014 Directive, whereas the ECJ reasoned and decided in relation to the 2004 Directive), both the AG Opinion and the Judgment are aligned in considering that the choice of one Directive over the other would have limited significance because the ‘definitions of “public contract” and “public work contracts” are equivalent in the two directives’ (Opinion, para 42) and ‘the content of Article 1(2)(b) of Directive 2004/18 corresponds in substance, as regards the execution of a work corresponding to the requirements expressed by the contracting authority, to the content of Article 2(1)(6)(c) of Directive 2014/24’ (Judgment, para 36).

However, this could mask disagreement on the (implicit) relevance of the new definition of procurement inserted in Art 1(2) of Directive 2014/24, which defines it as ‘the acquisition by means of a public contract of works, supplies or services by one or more contracting authorities from economic operators chosen by those contracting authorities, whether or not the works, supplies or services are intended for a public purpose’ (emphasis added). AG Campos explicitly reasoned in terms of the need for their to be an enforceable right to acquire the works (issue 1 below), whereas the ECJ decided not to use the words acquisition or acquire in its Judgment. This could signal a potentially problematic inconsistency in the interpretation of the extent to which the requirement for there to be an ‘acquisition’ modulates the scope of application of the procurement rules. This can be particularly relevant in relation to the delineation of the scope of application of the procurement and State aid rules, in particular in relation to the ‘de-risking’ of development projects, as further discussed below.

Issue 1: ‘acquisition’ and legally enforceable rights

As mentioned above, the first issue before the Court concerned the threshold to consider that a set or collection of agreements constitute an ‘acquisition’ and are thus covered by the scope of application of the EU public procurement rules, in particular where a contractor which is also a State aid beneficiary has a put option to transfer the works to the contracting authority.

In his Opinion, AG Campos provided a summary of the relevant case law (paras 52-54) and established that, ultimately,

… in order for there to be a genuine works contract, it is essential that the successful tenderer should specifically take on the obligation to carry out the works forming the subject of the acquisition and that that obligation should be legally enforceable. The contracting authority … must acquire the immovable property on which the works are carried out and, if necessary, take legal action to compel the tenderer awarded the contract to hand the property over to it, if it holds over the use of the works a legal right enabling it to ensure that they are made available to the public’ (para 60).

AG Campos had significant concerns about the way the factual pattern of the case had been presented to the ECJ. He made it explicit that ‘a reading of the order for reference and the subsequent course of the preliminary ruling proceedings [did not allow] to form a categorical opinion on the nature of the “collection of agreements” at issue’ (para 57), and pointed out at significant difficulties to determine what legally enforceable rights derived for the Slovak State, and that ‘it is not clear what performance the Slovak State may claim from NFŠ under the grant agreement and the agreement to enter into a future sales agreement, this being a premiss which it is for the referring court to determine’ (para 58). AG Campos also stressed that nothing in the written or oral submissions ‘support the inference that the Slovak State would have any right to take legal action against NFŠ to compel it to build the stadium should that undertaking ultimately decide not to do so. The difference is that, in that event, NFŠ would not have received the grant, or would have lost it, or would be obliged to pay it back. This in itself, however, has nothing to do with the performance of a works contract.’ (para 59), and that ‘all the indications are that the agreement to enter into a future sales agreement gave NFŠ the option either to remain the owner of the stadium and continue to operate it (or assign its operation to third parties), or to transfer it the Slovak State, if it suited it to do so’ (para 62).

This led AG Campos to conclude, on this issue, that

… there are many reservations to raise as against the classification of the “collection of agreements” at issue as a genuine public works contract within the meaning of Article 2(1)(6) of Directive 2014/24. Its classification as such or otherwise will be contingent upon the referring court’s final assessment of a number of factors informing the adjudication of the case which it has itself failed to mention with sufficient clarity (para 70).

The Court took a markedly different approach.

The ECJ considered that it ‘must take account, under the division of jurisdiction between the Court and the national courts, of the factual and legislative context, as described in the order for reference, in which the questions put to it are set’ (para 31). And, in relation to establishing the existence of the elements required for there to be a “public contract”, that ‘it will be for the referring court to rule on that matter, having made the relevant findings in that regard’ (para 39). This was probably to be expected and aligns with the general case law on the matter.

However, given the concerns on the lack of clarity of the evidentiary material before the ECJ, the absence of evidence of the existence of a legally enforceable obligation to build the stadium, the admission at the hearing by the Slovak Government that the put option was unilateral and discretionary (‘both NFŠ and the Ministry of Education expressed the same view in this regard, recognising that the (unilateral) option to sell was available for NFŠ to exercise if it wished to do so’ fn 44 in AG Opinion), and the broader indications, including in the State aid Decision, that there was no enforceable obligation against NFŠ because the exercise of the put option was entirely at its discretion, as stressed in the AG Opinion and as explicitly recognised by the ECJ too (‘Decision SA.46530 states that NFŠ will remain the owner of the Slovak national football stadium after its construction, without there being any obligation to transfer ownership of that stadium to the Slovak State’, para 58), the more specific reasoning of the ECJ is surprising.

The Court focuses in particular on whether the collection of contracts were concluded ‘for pecuniary interest’. It stresses that ‘the expression “for pecuniary interest” refers to a contract by which each of the parties undertakes to provide one form of consideration in exchange for another. The synallagmatic nature of the contract is thus an essential characteristic of a public contract, which necessarily results in the creation of legally binding obligations for each of the parties to the contract, the performance of which must be legally enforceable’ (para 44). This is another restatement of the case law and, given the framing of the issues above, one would have expected the ECJ to stress at this point that the referring court is the one that needs to establish whether there are such legally enforceable obligations, perhaps stressing the elements that question such a finding as laid out in the AG Opinion.

This is not what the ECJ wrote in its Judgment. The Court said

… where a contract includes an obligation to purchase by a contracting authority without an obligation to sell devolving on the other contracting party, that absence of an obligation to sell is not necessarily sufficient to rule out the synallagmatic nature of that contract and, therefore, the existence of a public contract, since such a conclusion may, as the case may be, be reached only after an examination of all the relevant factors para 45, emphasis added).

In the present case, the referring court mentions the existence of reciprocal obligations between the Ministry of Education and NFŠ. In addition, that court states, inter alia, that the grant agreement imposes an obligation on the State to award the grant and the obligations [for NFŠ] to construct the Slovak national football stadium in accordance with the conditions specified by the Ministry of Education, to finance at least 60% of the construction costs … (para 46).

…

a collection of agreements binding a Member State to an economic operator and including a grant agreement and an undertaking to purchase, concluded with a view to building a football stadium, constitutes a ‘public works contract’ within the meaning of that provision, where that collection of agreements creates reciprocal obligations between that State and that economic operator, which include the obligation to construct that stadium in accordance with the conditions specified by that State and a unilateral option in favour of that economic operator corresponding to an obligation on the part of that State to purchase that stadium, and grants the same economic operator State aid recognised by the Commission as being compatible with the internal market (para 61, emphasis added).

Crucially, this conclusion of the ECJ fails to explicitly stress that ‘It is for the referring court to determine whether those circumstances are present in this case’, which the AG Opinion did include (para 96). Although the Court does mention in passing that its considerations are based on elements that are ‘subject to the verifications to be carried out by the referring court’ (para 55), by not making this explicit in the answer to the question, the ECJ raises significant questions and potential difficulties once the litigation proceeds at the domestic level.

It is also notable that the ECJ, despite fundamentally saying the same as the AG once it is clear that all relevant findings of fact and their legal implications need to be ascertained at domestic level, chose to phrase its overall conclusion as the opposite default as AG Campos.

AG Campos had proposed that the Court should find that the relevant rules

must be interpreted as meaning that a grant agreement and an agreement to enter into a future sales agreement which are concluded between a State body and a private undertaking and in which the private undertaking is granted public funds for the purpose of the construction of a sports infrastructure and is given the unilateral option of selling it to the State, respectively, cannot be classified as a public works contract if they do not give rise to a legally enforceable obligation for the State to purchase the infrastructure and if the State does not derive a direct economic benefit or has not had a decisive influence on the design of the work. It is for the referring court to determine whether those circumstances are present in this case.

This formulation created the default rule that put options are not presumptively covered by the procurement rules, and stressed the need for the domestic court to positively find application of the three cumulative criteria determinative of an acquisition covered by the procurement rules (enforceable obligations, direct economic benefit and decisive influence in the design).

Conversely, as mentioned above, at para 61 the Court found that the concept of ‘public works contract’  extended to a ‘collection of agreements creates reciprocal obligations between that State and that economic operator, which include the obligation to construct that stadium in accordance with the conditions specified by that State and a unilateral option in favour of that economic operator corresponding to an obligation on the part of that State to purchase that stadium …’.

This can create the impression that put options are presumptively covered by the procurement rules. However, in my view, this would not be an adequate reading of the case. For three reasons.

First, because the answer given by the Court in relation to the enforceable obligations is in part tainted by its failure to stress that this is subject to verification (as above).

Second, because the ECJ also made quite a peculiar distinction between the presumed obligation to build the stadium and the discretionality of the put option when it stressed, in relation to the State aid Decision, that ‘although Decision SA.46530 states that NFŠ will remain the owner of the Slovak national football stadium after its construction, without there being any obligation to transfer ownership of that stadium to the Slovak State, that decision does not mention the absence of an obligation to construct that stadium’ (para 58, emphasis added). This strongly suggests that the answer of the Court is primarily focused on the presumed obligation to build the stadium.

Third, because the ECJ’s approach to assessing the extent to which a put option creates a direct economic benefit for the contracting authority also raises some questions, as discussed below.

Issue 1: Direct economic benefit

An issue that had not featured prominently in AG Campos’ Opinion is whether the “collection of agreements” would have been to the direct economic benefit of the contracting authority. The Opinion simply stressed that it was unclear whether ‘the Slovak State obtained a direct economic benefit from the two agreements at issue … The State’s interest (and subsequent indirect benefit) seems to be confined to the generic promotion of the national sport’ (para 64).

By contrast, the Court engaged in a more detailed discussion, which it is worth reflecting in full:

… in a public works contract, the contracting authority receives a service consisting of the realisation of works which it seeks to obtain and which has a direct economic benefit for it. Such an economic benefit may be established not only where it is provided that the contracting authority is to become owner of the works or work which is the subject of the contract, but also in other situations, in particular where it is provided that the contracting authority is to hold the legal right over the use of those works, in order that they can be made available to the public.

It is apparent from the documents before the Court that, although the Slovak national football stadium belongs to NFŠ, the grant agreement limits the right to transfer ownership of that stadium to third parties, in particular by requiring prior written consent from the Slovak State in order to do so. Therefore, that State has, with regard to this stadium, in essence, a right of pre-emption with an intrinsic economic value.

The economic benefit may also lie in the economic advantages which the contracting authority may derive from the future use or transfer of the work, in the fact that it contributed financially to the realisation of the work, or in the assumption of the risks were the work to be an economic failure (see … Helmut Müller, C‑451/08, EU:C:2010:168, paragraph 52 and the case-law cited).

In the present case, as NFŠ stated in its written observations and at the hearing, the option available to it under the undertaking to purchase constitutes a guarantee against the commercial risk in the event that the Slovak national football stadium proves to be commercially unviable for it. Thus, by undertaking to purchase that stadium at the request of NFŠ, the contracting authority assumed all the risks were the work to be an economic failure (paras 47-50, emphases added).

There are two points worth discussing here. The first one concerns the pre-emption right. The second one concerns the issue of the assumption of risks. Both are relevant from the perspective of the interaction between State aid and procurement law.

First, a right of written authorization for a transfer does not amount to a pre-emption right. The State could have the right to veto a transfer without this giving it priority to acquire the asset. It could simply be that the State has the right to screen for a suitable owner of the stadium, but that the legal consequences of denying the authorization do not immediately amount to the right to acquire instead of the proposed buyer. Rejection of authorisation may solely result in NFŠ having to put forward an alternative buyer, or deciding to keep the stadium. Moreover, the ECJ does not engage in the possible logic of the pre-emption right from an economic viewpoint, which can have more to do with the State’s interest in having a say over the transfer of the stadium in potentially heavily subsidised conditions, eg to ensure that there is no circumvention of relevant sets of fiscal rules, than in relation to a potential direct acquisition of the stadium. An absence of any such reasoning by the ECJ raises significant questions on the treatment of a (presumed) pre-emption right as a direct economic benefit.

Second, the way the Court engages with Helmut Müller is in itself problematic. Not least because there seems to have been a deformation of the ‘Auroux formula’ as it has migrated through the case law of the Court. It is worth recalling that Auroux (C-220/05, EU:C:2007:31) concerned a case involving the signing of an agreement between a municipality and a special purpose vehicle with separate balance sheet to run a re-generation programme. That re-generation programme expected to make profits from the sale of real estate to third parties. The agreement foresaw that, at the end of the project, ‘Any excess on that balance sheet is to be paid to the municipality. Furthermore, the municipality automatically becomes owner of all the land and works to be transferred to third parties not yet sold’ (para 18). This is the context in which the ‘Auroux formula’ as enunciated in Helmut Müller needs to be understood. Nothing in Helmut Müller itself questions the proper understanding that there has to be a direct positive economic benefit arising for the contracting authority—if anything, the opposite is true.

However, in NFŠ, paras 49 and 50 of the Judgment seem to suggest that ‘the assumption of the risks were the work to be an economic failure’ can in itself amount to an economic benefit. This makes no plain sense, as the assumption of such risks is clearly an economic disbenefit or liability for the State. Moreover, it does not make sense in the context of Auroux itself, where the economic benefit consisted of ‘the economic advantages which the contracting authority may derive from the future use or transfer of the work’, as the municipality was indeed entitled to potential profits of the sales to third parties, as well as in line to immediately acquire any unsold real estate. The reason why the municipality could obtain such benefits or, in other words, the consideration given to the developer consisted in its financing and the de-risking the project—but that did not turn the financing or de-risking themselves into economic benefits!

It is thus important to stress that the State has to derive a positive economic benefit or advantage, such as sharing in the revenues of the transfer of assets, or getting to use them. In the NFŠ case, the Slovak State would neither participate in the proceeds from the sale of the stadium to a third party, nor have the right to use the stadium. Quite which economic benefit the Court identified is thus also unclear—if not plainly incorrect. This is important from the perspective of the substantive interaction of procurement and State aid rules, especially bearing in mind that State aid related to infrastructure tends to imply a mix of measures concerning the financing and de-risking of development projects. If taking risks was by itself to be considered as obtaining an economic advantage, the potential subjection of a significant number of State aid measures to procurement would be a clear risk. It would, however, be at odds with the general approach of Directive 2014/24/EU. We should not lose sight from the fact that, as AG Campos stressed in his Opinion ‘the mere grant of a State subsidy involving the movement of public funds … does not in itself amount to the conclusion of a public works contract. As recital 4 of Directive 2014/24 states, “the Union rules on public procurement are not intended to cover all forms of disbursement of public funds, but only those aimed at the acquisition of works, supplies or services for consideration by means of a public contract”’ (para 48, emphasis in the original). By the same token, not all forms of de-risking of infrastructure projects are necessarily covered by public procurement law.

Issue 2: Prior approval of the State aid measure

The second relevant issue on which the Judgment could have provided clarity concerns the extent to which the prior approval by the European Commission of a State aid measure explicitly detailing a strategy to comply with EU public procurement law should bind future assessments of compliance with those rules. In that regard, the Opinion had been clear and ambitious, when AG Campos stated that

The Commission can actively intervene in defence of competition where public procurement does not comply with the rules laid down in, inter alia, Directive 2014/24 in order to safeguard this objective. I do not see any reason why it should not do so when faced with an examination of the viability of State aid measures resulting from agreements concluded by public authorities with private entities.

In particular, it is my view that the Commission could not have failed to examine whether the form in which the public aid granted to NFŠ was structured masked the existence of a public contract which should have been put out to tender. To my mind, it did so implicitly, which explains paragraph 8 of its Decision SA.46530.

In short, Decision SA.46530 is based on the premiss that there was no obligation to transfer ownership of the stadium to the Slovak Republic. That assumption, to which I have already referred, cannot be called into question by the referring court, which must respect the Commission’s assessment of the factors determining the existence of State aid (paras 77-79, emphasis added but underlined emphasis in the original).

By contrast, the ECJ fudged the issue by stating that

… it should be noted that it is true that national courts must refrain from taking decisions running counter to a Commission decision on the compatibility of State aid with the internal market, the assessment of which falls within the exclusive competence of that institution, subject to review by the Courts of the European Union … However, assessments which might implicitly follow from a decision of that institution relating to State aid cannot, in principle, be binding on the national courts in a dispute, such as that in the main proceedings, which is unrelated to the compatibility of that aid with the internal market (para 59, emphasis added).

This deserves some comments.

First, the suggestion by the ECJ that the fact that the assessment of the compatibility with EU law would arise only implicitly from the State aid decision and thus could not be relied on is problematic. Mainly, because it is at odds with previous case law and, in particular, with the position that the Commission can discharge its obligations to assess State aid measure’s compatibility with other fundamental provisions of EU internal market law, including secondary EU law, by implication. For example, in Castelnou Energía, the General Court accepted that the consideration of those rules can be implicit if the reasoning of the Commission refers to those other rules of secondary EU law and they feature in its analysis (T-57/11, EU:T:2014:1021, at para 185). Therefore, an implicit assessment would suffice where compliance with EU procurement rules include a reference to those rules and it features in the Commission’s State aid analysis. This was the case in NFŠ, where the Commission had explicitly stated that ‘The construction works financed through the grant … will be subject to a competitive process, respecting the applicable procurement rules’ (SA.46530, at para 8).

Second, this statement comes to create problems in domestic litigation where an argument is made that a dispute in a case concerning State aid concerns issues ‘unrelated to the compatibility of that aid with the internal market’, as it will many times be the case that compatibility is not primary reason why the measure is challenged, but the Commission will have taken it into account in its assessment. If anything, limiting the bindingness of Commission State aid decisions in this way erodes the monopoly of application of State aid rules given to the Commission in Art 108(3) TFEU.

Final thoughts: obiter dicta?

The analysis above has hopefully shown how the NFŠ Judgment can be problematic. However, I submit that, on a proper interpretation of the case and relevant precedent in their circumstances, most of the problematic statements need to be taken as obiter dicta because they are not backed by the facts of the case and, therefore, constitute general statements made in passing by the Court that cannot alter the relevant position of these issues under EU law.

First, I have highlighted how it is problematic for the NFŠ Judgment to suggest that put options are presumptively covered by the procurement rules (para 61). This is because such suggestion is in reality mixed up with a presumption of an obligation to build the infrastructure over which (at the very least) significant questions loom large. To me, it seems clear that the Judgment accepts that it is not the position under EU procurement law that a purely unilateral option to sell that is not enforceable by the contracting authority does not meet the requirement to establish legal obligations. However, the formulation used by the ECJ and the omission of the precision that establishing whether any legal obligations were created in the case is for the national courts, is confusing in this regard.

Second, and still on the issue of NFŠ’s transfer rights, I have also highlighted how the suggestion that a requirement for written authorisation of a sale to a third party implies a pre-emption right that has intrinsic economic value (para 48) is also problematic. On this, much more detailed legal analysis of the specific content of rights arising from the requirement for such authorization would be required. And, once again, this would be for the national courts.

Third, I have highlighted how a maximalistic and de-contextualised approach to understanding that de-risking infrastructure projects (para 50) could in itself constitute an economic benefit would also very problematic. I have suggested that a proper understanding of the ‘Auroux formula’ as enunciated in Helmut Müller must always imply the existence of a positive economic benefit, and that it cannot be conflated with the disbenefit or liability accepted by the contracting authority or State aid grantor as potential consideration for such (future) economic benefit.

Finally, I have highlighted how the suggestion that implicit assessments of compatibility with EU procurement law contained in State aid decisions cannot be relied on (para 59) is also problematic and at odds with existing case law. More generally, a partitioning or limitation of the types of disputes over which a Commission State aid decision has binding effects is undesirable.

How to get out of these potential problems, then?

The way forward requires paying close attention to the circumstances of the NFŠ case.

On the first issue, the ECJ itself was clear that the Commission had accepted that ‘Decision SA.46530 states that NFŠ will remain the owner of the Slovak national football stadium after its construction, without there being any obligation to transfer ownership of that stadium to the Slovak State’ (para 58) and the AG had documented that ‘both NFŠ and the Ministry of Education expressed the same view in this regard, recognising that the (unilateral) option to sell was available for NFŠ to exercise if it wished to do so’ (AG at fn 44). It is thus not in dispute that the put option did not create any legally enforceable obligation. Therefore, a suggestion that a put option could presumptively create legal obligations and thus be caught by the procurement rules has no relation to the facts of the case and needs to be taken as obiter dictum.

In NFŠ, the core obligation the Court takes issue with concerns the primary obligation to build the stadium. However, on that issue, even if not clearly, the ECJ has not deviated from the EU law position that ascertaining the existence of legal obligations is a matter for the domestic courts (para 31).

The second issue goes away on the basis of the same principle. Simply put, the ECJ has no jurisdiction to assess that by virtue of NFŠ’s obligation to require prior written consent from the Slovak State to transfer ownership of that stadium to third parties ‘that State has, with regard to this stadium, in essence, a right of pre-emption with an intrinsic economic value’ (para 48). This is a matter for the national courts and, consequently and at most, the ECJ statement can only be seen as an obiter dictum.

The third issue also concerns an obiter dictum approach by the Court. At its core, the Auroux line of case law is irrelevant to NFŠ to the extent that both cases can be clearly distinguished. In Auroux, the contracting authority was in line to share in the above agreed balance sheet benefits and/or to acquire unsold real estate. In NFŠ, there was no right to participate in the future transfer of the stadium to third parties. Therefore, all other statements as to how the precedent would apply to the case hand if the case at hand was different must also be considered an obiter dictum.

Finally, the position that implicit assessments of compatibility with EU law in State aid decisions cannot be relied on in relation to disputes about anything other than the compatibility of the aid is also not of relevance of the case because, in reality, the “collection of agreements” constituted the State aid measure and challenging it for breach of fundamental rules of internal market law is nothing else than challenging its compatibility with the internal market. Therefore, this statement is also obiter dictum.

Overall, it seems to me that the NFŠ Judgment is problematic in the ways in fails to provide clarity on the interaction between State aid control and public procurement law. At the same time, its legal value is limited because it does not really deviate from established precedent and, in the areas where it would suggest it does, it would do so in deviation from the facts of the case at hand. It is regrettable that the Court decided not to follow the much clearer and productive proposals advanced by AG Campos in this instance.

Analysing the potential pitfalls of “AI-judication” in Public Procurement -- guest post by Paris & Esmail

This guest post by Dr Matthew Paris* and Dr Zack Esmail** complements some of the earlier guests posts on AI use in the public sector by focusing on the use in public procurement and, in particular, in the review of procurement decisions.

I found their point about the potential futility of review and appeal procedures underpinned by the AI that was used during the original procurement process particularly thought-provoking. Happy reading. And, if you feel inspired and want to submit your own post, please do get in touch: a.sanchez-graells@bristol.ac.uk.

Rens Dimmendaal & Johann Siemens / Better Images of AI / Decision Tree reversed / CC-BY 4.0.

Computer Says No?
Analysing the potential pitfalls of “AI-judication” in Public Procurement

Introduction

The introduction of new technologies have always thrust the issue of Public Procurement into the spotlight, subtly or otherwise. Recently, the Overton window in tech circles has been dominated by one discussion uber alles: Artificial Intelligence.

So far, AI has already obtained credible influence in the way public procurement tenders can be drafted and submitted. Over the next few years, many anticipate larger strides, in hopes of a more efficient and smooth procedure for any individual or entity interested in a bidding procedure for a public service or product.

However, this technology has not yet registered a significant foothold when it comes to the evaluation process. Indeed, unlike other elements related to the monitoring and effectiveness of spending control, the issue of using AI tools contain its own legal and ethical quandaries. The answers for which remain not only elusive in practice, but are the subject of considerable debate in theory.  Approaching this debate is no longer an option – it’s becoming more of a necessity.

Qui custodiet ipsos custodes?

The apex of any public procurement procedure is in its decision stage – be it through a sole individual, a tribunal or a committee. The forms and procedures may differ between countries, but the crucial elements of transparency, expediency and adherence to regulations, remain global.

The adjudicators or evaluators of a tender act as both gatekeepers on behalf of the tendering process. Simultaneously, they are also entrusted by the State within the Executive, in ensuring that the process remains smooth and cost-effective.

This process, at present, is not without its faults. Evaluation Committees have many a time been deemed to be improperly constituted, due to cases of prima facie impartiality, or conflicts of interest. In other instances, evaluation organs have acted ultra vires, or were found at fault for any substantive and/or procedural requirements alike. This in itself has led to inter alia an increase in costs by supplier and Government alike, significant delays, and further discontent in national project delays.

Aside from the identifiable characteristics, there is also the probability of implicit biases (such as triggering personal experiences, instances of groupthink, political and ideological agendas,) that members of an evaluation committee, unintentionally or otherwise, fail to address properly.

These are issues inherent in human nature. And yet, most of the time, they play a subtle yet important role in anyone’s decision making. As an end result, evaluators unknowingly or otherwise, adjudicate with the lack of true objectivity so fundamental to any evaluation that takes place.

Consistent Criteria, or Intransigent Checklisting?

Proponents of Artificial Intelligence point towards the above issues as further ailments of a system in need of a revamp. Furthermore, another point argued specifically by those with some experience of tender evaluation within their local Governments is that, more often than not, such an adjudication ends up more akin to a checklist exercise.

=IF A, THEN B.

Such an inflexible state of affairs may do well in the bulk of tendering processes. This is not exclusive to the least economically significant of procurement – many elements related to tendering processes do indeed require formulaic, standardised requirements, which help ensure compatibility with extraneous political, legal and environmental constraints. Indeed, any type of software, especially one as intelligent as AI, can alleviate considerable pressures in ensuring such conformity, and leave room for even more ambitious projects to materialise.

However, adherence to standardised templates can only lead one so far.

A tick the box exercise

Current AI technology harvests (or is fed) substantial amounts of data, with the ability to process and regenerate the requested output. On paper, this procedure allows any user to obtain a result which is not only optimal, but also consistent. And for many proponents within the public sector, this element of procedural and substantive uniformity, is one of the cardinal rules by which all tendering procedures should abide by.

However, no public procurement department operates in isolation. Any evaluation process must keep in mind issues such as market forces and supply issues, local and/or regional. This is all the more relevant for small-scale economies.

Of course, an evaluation committee made out of human beings, especially those intimately familiar with these technicalities, have the ability to calculate these issues, and decide on the fairest way forward. Any evaluation process that takes place in the adjudicatory stage of any public procurement process, must be in tune with all relevant matrices necessary for an informed decision. For human evaluators, this information can be brought to their attention by either the bidders themselves, or through the research that they themselves conduct when analysing the submissions.

For any AI software, this relevant data (ironically) may not be fully automated. Rather, not only must it be submitted as part of its evaluation criterion, but rather prioritised accordingly to its importance in the relevant geo-political context. If not, then the whole tendering process would be built, evaluated and submitted for approval in a manner wholly unrepresentative of the scenario outside the boardroom.

What this may mean is a zero-sum game, with all parties facing a litany of different issues: an ineffective service or product, which fails to reach any of its intended targets; a public body facing risky cost overruns, and an internal or external outcry for failing to delivery on what was promised; a bidder who will either refrain from engaging with other public bodies in the future, or else act comfortably enough within his assigned position to demand payments or conditions over and above those agreed upon and stipulated upon; and the taxpayer being left with a hefty bill for marginal gains, if any.

The Appellate Stage – More than just Ones and Zeros?

The current human element when it comes to adjudication in the first instance, is also important if and when we reach the appellate stage.

Appeals can be lodged for many reasons, as discussed in a previous section. And to the trained eye, this is par for the course: A legal system working as designed, ensuring as much as possible that the most equitable and just conclusion is reached the second time around.

Another way to look at it is as follows: An appeal is usually the least direct outcome. Furthermore, the possible use of AI in appellate stages, definitely leads to more questions than answers.

Philosophically, the use of AI software can lead to a political and legal Catch-22 for any entity or evaluator. Let’s say that such a technology is applied in all evaluation stages: Both in the first instance, and in subsequent re-evaluations. Should the same AI software used in the first stage of proceedings, be used again in the second stage? Will the same parameters used in the initial evaluation, remain unchanged? If that were to occur, then any subsequent requests for review will be inherently worthless. Expecting a different outcome from the same set of variables over multiple times is not a process of adjudication – it’s an exercise of ticking a box. If the AI software in the second stage of proceedings is indeed operable on a different dataset, or a different software entirely, then another question pops up: What parameters are changed? And if the evaluation software is entirely different, what justification is there for one system to be held on a higher legal level than the other?

One may answer by requiring that such a software plays only a limited part in adjudicating any tenders or calls issued by a Government entity. If the software is strictly used in the first instance, then the essential human element in the second instance not only remains – it supersedes that of the technology, the same software initially sought for to remove any issues brought by human interference.

If the opposite scenario plays out, then it’s the human element which is subservient to the rigidity of such a software. This however, downgrades the human ability of foresight and contextualisation to that of a program, leaving all decisions at the mercy of an uninvolved, absent and uninterested party – the software programmer.

Concluding comments

All things must change, and in turn so must we. The technostructure affecting our daily lives continues to be transformed at breakneck speed – in turn, Governments, businesses and other associations are struggling to remain in step with the times, let alone ahead of them.

Indeed, while a discussion on the role of AI seemed merely the work of science-fiction a few years ago, it is nowadays a prominent feature. More so in an area of public administration so tightly regulated, a sphere of law so frequently the source of public controversy and media speculation. However, the influence that this technology has had, and may continue to have, on our daily lives as citizens, is rapidly accelerating.

Whatever role that AI may play in the next few generations, is yet to be fully deciphered. Alas, this technology must never supplant the fundamental rights and principles that have ensured faith, trust and accountability in all public procurement proceedings. If that Rubicon is crossed, and this trust is disintegrated, the involvement of both private entities and individuals withers away, paralysing public entities everywhere, and removing the stability so pivotal that is provided for the good of all citizens alike.

Dr. Matthew Paris

Dr Matthew Paris is an advocate and Founding Partner of DalliParis Advocates, a boutique law firm in Malta. Renowned for his expertise in public procurement and corporate law, Dr Paris has successfully represented clients in domestic public procurement disputes, as well as other complex disputes involving an international dimension.

Dr. Zack Esmail

Dr Zack Esmail is an advocate and Associate at DalliParis Advocates, specialising in Commercial Law, Public Procurement and Intellectual Property. He graduated with a Masters of Advocacy in 2023 after defending his dissertation entitled ''An analysis of the notion of public interest in private projects from a human rights perspective''

Public procurement of AI between opportunities and risks -- guest post by Giuseppe Bitti

This guest post by Giuseppe Bitti* explores the broad policy approaches that can be followed to slow down AI adoption by the public sector, with a focus on risk mitigation and management.

It was submitted to the ‘a book for a blog’ competition and has won Giuseppe a copy that will soon be on the post.

Please note Giuseppe’s disclaimer that “The views expressed are those of the author and do not necessarily reflect those of the European Central Bank”.

Catherine Breslin & Rens Dimmendaal / Better Images of AI / Strawberries and milk / CC-BY 4.0

Public procurement of AI
between opportunities and risks

Introduction

In the past year, the growth in hype around AI and machine learning has been considerable. The thesis that most white-collar tasks will be automated within a few years (or even months!) would seem (for many) to have dispensed with the requirement to be subject to empirical demonstration, having thereby morphed into an item of faith.

IT providers are naturally keen to promote their AI solutions – claiming they can cover most of their customers’ needs – and often AI and machine learning are mixed with standard digitalisation, leading to additional confusion. Furthermore, this new AI ‘gold rush’ may lead to acritical applications of AI tools to inadequate use cases and situations where it might do more harm than good, out of FOMO by the said customers.

While the potential benefits are clearly appealing and AI will most likely have a strong impact on how a lot of white-collar tasks are performed – potentially revolutionising many of them – it would be wise to take this hype with a pinch of salt and proceed cautiously, especially in areas which are particularly sensitive, such the exercise of public tasks and powers, and the management of public money and resources (including data).

As known, the EU has taken action along this lines and has been (again) leading worldwide with the first general legal framework for AI, defining i.a. categories of tasks for which AI can be freely used, or used subject to light requirements (e.g. transparency for image manipulation), or substantial requirements (e.g. for law enforcement and essential public services), or not used at all (e.g. facial recognition via CCTV footage scraping).

However, besides general legislation, the aim of a pondered adoption of AI by the public sector can also be achieved via other, specific means, including public procurement, especially considering that the latter will likely be the main ‘entrance gate’ for AI tools into the public sector, except for those (very) few public administrations sufficiently large (and deep-pocketed) to be able to develop their own in-house solutions.

In this regard, it is noteworthy that the AI Act entrusts the Commission (and specifically the EU’s AI Office) with the task of “evaluating and promoting the convergence of best practices in public procurement procedures in relation to AI systems” [Article 62(3)(d) EU AI Act].

At first glance it might not appear very easy to use public procurement as a tool to delay other public sector activities (namely, adoption of AI tools). Public procurement is generally meant to be a tool for enhancing – and not hindering – the pursuit of public sector activities.

However, it is also true that public procurement can – and probably should – be a tool for the safeguard and promotion of interests other than the traditional ones. A good example of this is the use of public procurement to contribute to the promotion of sustainability objectives, even to the detriment of traditional aims such as ensuring the best value for money in a strict, financial sense.

Hence, we could also frame the task of using public procurement to slow down AI adoption by the public sector as a contribution of public procurement policy to the (desirable) aim of a cautious and thought-through adoption of AI tools in the public sector, even if to the detriment of the alternative aim of a swift – but risky – reaping of the appealing benefits offered by AI tools.

Some options to use public procurement to slow down AI adoption in the public sector

As mentioned above, public procurement should generally foster the action of public administration. However, it might also be used to steer it towards complementary dimensions. If so, what is to be done?

On a (trite) humoristic note: a first idea to slow any task down would be to leave its implementation to a cumbersome public tender procedure.

Dull jokes aside, I think several options can be identified, below in order of ‘invasiveness’.

Ban or Moratorium

A first, draconian idea would be to (temporarily) prohibit contractors from relying on AI tools for the provision of services to the contracting authority. This could be difficult to implement, especially as an increasing number of suppliers do rely – and will rely even more – on AI solutions for all their clients.

Still, public procurement does account for approx. 14% of the EU’s GDP, hence the margin for a strong suasion is there.

The first, direct result of this measure would be – self-evidently – to slow down the explicit use of AI solutions by the public sector.

However, such a measure would also likely contribute to an indirect slowing down of general AI adoption, by decreasing the providers’ overall economies of scale in adopting AI solutions.

A supplier might need to think again whether it makes sense to change its delivery model to include AI tools if this would benefit only its private clients, while it would need to keep the current non-AI-based delivery model towards its public sector portfolio.

AI-keen suppliers might not bid for public procurement opportunities, losing an important revenue stream, while other contractors might be forced to keep up two parallel systems, making AI adoption more costly due to reduced economies of scale.

Suppliers working predominantly or exclusively for public sector clients might decide to delay the adoption of any AI solution as it would expose them to more risks (e.g.: exclusion from the procurement procedure, contract termination) than potential benefits (namely prevailing in procurement procedures over competitors which would also be prevented from using AI).

Neutralising Risks

A second idea would be to follow a similar logic to that sometimes applied in sustainable procurement and focus on neutralising the main risks of AI.

For example, if the main concern is to avoid falling for the hype of an appealing solution promising extraordinary results but also exposing to a considerable financial risk due to possible mistakes made by the tool, such risk could be tackled in part via tailored tender specifications.

These could for example foresee penalties for gross (or even minor) errors made by the offered solution. To avoid later calculation issues, such liability could be precisely stipulated and agreed in the form of a service level agreement, foreseeing fixed amounts per category of issue/error.

This idea could also be complemented by a requirement – at selection stage – for the contractor to provide a financial guarantee (e.g. an insurance).

This would first ensure the solvency of the contractor (which might also not be a large AI player, but a simple software reseller, or a company relying on AI provided by another subcontractor).

Secondly, it would require the involvement of a third-party guarantor/insurer, thereby increasing the level of scrutiny on the precise functioning (and risks) of the AI solution and generally increasing the costs and complexity of providing it, especially in case of use by a ‘normal’ supplier – e.g. a provider of facility management services – of AI tools provided by a subcontractor.

Weighing Risks

A third idea, similar to the one above, would be to focus on weighting the main risks presented by the AI tool by enhancing their relevance in the technical evaluation of the offers.

Of course, assessing the risks of the proposed solutions is already standard practice in all procurement processes, especially for IT tools.

However, the weighting of such criterion could be enhanced beyond usual standards to increase the relevance of risks presented by AI solutions vis-à-vis other technologies. As mentioned, in a lot of cases, recourse to AI is a possible, but not a necessary solution to a genuine need for digitalisation.

An appropriate weighting of the risk factor could help to separate the two needs (actual vs potential need for AI). Suppliers might spontaneously decide not to offer an AI tool, if this exposed them to the unnecessary risk of a negative assessment.

Complementary measures to risk neutralisation and risk weighing

These options should be complemented by additional safeguards, which might also result in a slowdown of AI adoption.

Focus be either on risk neutralisation or weighing, the proposed solution must be thoroughly evaluated in the context of the procurement process as part of the offers. This could be done ideally via targeted functionality tests (e.g. proof-of-concept).

However, a known difficulty in algorithmic evaluation is the large variety of use cases to which AI can be applied. Therefore the contracting authority should be precise in identifying the specific use cases for the tool so as to allow for a meaningful evaluation.

Precisely defined benchmarks and use cases will also help with the auditing of the tool at a later stage. A recurrent auditing is particularly important, as the performance of the tool will evolve due its capacity to learn (and sometimes possibly even detect it is being tested).

Furthermore, in any case both the technical and the governance dimensions should be considered: the focus should not be limited to assessing the technical functionalities of the solution or its capacity to perform in a series of targeted tests, but also the adequacy of the set of control and quality management mechanisms implemented by the provider.

The contracting authority should also ensure that the contractor remains responsible and accountable for the output of its AI solution: the risk of a public administration relying on a tool which is too complex to understand both for the customer and for the contractor providing it – especially when via subcontracting – is too large to be neglected. Besides penalties, contractors should remain liable for damages caused by the solution they provide.

Finally, specific requirements should limit the geographical scope where the AI tool provider is allowed to process data, especially personal and confidential data.

Risks of a slowdown in AI adoption in the public sector

The aim of using public procurement to slow down AI adoption in the public sector has several merits as mentioned above, but it is not risk-free. Especially the option of a general ban/moratorium on procurement of AI tools (Option 1).

The main risk is that service provision by the public administration may remain ‘stuck in the past’, i.e. relying on outdated technologies and processes. In the long run this would become a structural weakness in the capacity of the public sector to deliver on its mandate and stay competitive.

Furthermore, a considerably more efficient private sector – thanks to adequate AI adoption – would naturally result in a stronger case for outsourcing those public tasks which could benefit from such advantage, with all the known risks related to loss of control by the public sector.

Finally, being a market player – as AI customer – also allows contributing to the development of both the market and product itself. Withdrawing completely – e.g. via a moratorium – would result in a development of the AI market which is neither shaped nor affected by the needs and concerns of the public sector, inevitably leading to a later need to use – or work on the adjustment of – solutions originally developed for another kind of clients, often with very different needs.

In conclusion, for most tasks performed by the public sector, the best options would likely be those aiming at neutralising (Option 2) or weighing (Option 3) the actual risks of AI, rather than prohibiting it altogether.

However, this is no size-fits-all: several core tasks performed by the public sector where the AI-related risks exceed the respective benefits would probably be better off by a freeze until additional safeguards can be deployed. This could be the case, for example, for AI tools contributing to key administrative decision-making (e.g. issuing a key license/permit or awarding a sensitive contract).

Ultimately it might be preferable to perform some tasks in a way which is less efficient, yet also less risky and possibly more ethically acceptable. Furthermore, for such tasks there would be less/no public sector competitivity concerns, due to the lack of competition.

However, in such cases the best fitting measure would rather be a moratorium on AI adoption via general legislation on the lines of the AI Act, rather than via public procurement policy.

Giuseppe Bitti

Giuseppe Bitti is a Procurement Expert in the Directorate Finance of the European Central Bank (ECB). Before joining the ECB, Giuseppe was Adjunct Lecturer of European Economic Law at the Faculty of Law of the University of Hamburg. His research interests include public procurement law and governance, AI procurement, and the impact of AI on public procurement processes. You can connect with him on LinkedIn.

A good excuse? Delaying the Procurement Act 2023 to make procurement mission-oriented

The UK has been in a long process of procurement law reform that was due to kick in on 28 October 2024, following a 6-month ‘go live’ notice for the Procurement Act 2023. Today, a 4-month delay has been announced and the new UK procurement rules will only enter into force on 24 February 2025 (barring any further delays).

This announcement follows speculation (on LinkedIn) that a delay was necessary because the digital platform underpinning the new system would not have been ready on time. However, the reasons given by Cabinet Office point in a different direction. In the statement, the UK Government has indicated that, under the Procurement Act 2023,

… the previous administration (the Conservative government) published a National Procurement Policy Statement [this one] to which contracting authorities will have to have regard. But this Statement does not meet the challenge of applying the full potential of public procurement to deliver value for money, economic growth, and social value. [The Government has] therefore taken the decision to begin the vital work of producing a new National Procurement Policy Statement that clearly sets out this Government’s priorities for public procurement in support of our missions.

It is crucial that the new regime in the Procurement Act goes live with a bold and ambitious Statement that drives delivery of the Government’s missions, and therefore, I am proposing a short delay to the commencement of the Act to February 2025 so this work can be completed.

To be honest, I do not find this justification very convincing — whether it is a curtain to cover problems with the digital platform or not. This is because the National Procurement Policy Statement (NPPS) is bound to play a very minor role in procurement practice, given the awkward position it has within the set of duties arising from the Procurement Act 2023—as I explore in detail here. Moreover, the NPPS is by design bound to change in the future (especially when there is a change of government), but future changes to the NPPS will not trigger the sort of pause that is being implemented now. And, finally, pushing back the entry into force of the Procurement Act 2023 and the associated NPPS leaves is in the current default, where the previous iteration of the NPPS (this one, also by a Conservative government—I know, confusing, so let’s call it the ‘old’ NPPS) is supposed to be guiding procurement decisions. So, stopping the entry into force of the new NPPS to leave the old NPPS is not precisely going to create any change in policy and practice between now and February 2025 (even assuming the NPPS has policy delivery potential).

For me, this move and delay is concerning because it shows how the incoming Government places excessive hopes on procurement, and the NPPS in particular, as a policy delivery tool. More than ever, this stresses the need to have a serious conversation about the limits of procurement law and the need for hard law in many areas (starting with tackling climate change and supporting a just environmental transition) if we are to unlock the change we need at the required pace.

Articulating the public interest in procurement law and policy

Earlier this year, I had some very interesting conversations with Bristol colleagues about the relationship between law, regulation, and the public interest. These conversations led to a series of blog posts that are being published in our Law School blog.

This prompted me to think a bit more in detail about how the public interest is articulated in public procurement law and policy. Eventually, I wrote a draft paper based on the review of procurement goals embedded in the UK’s new Procurement Act 2023, which will enter into force next month (on 28 October 2023).

I presented the paper at the SLS conference today (the slides are available here) and had some initial positive feedback. I would be interested in additional feedback before I submit it for peer review.

As always, comments warmly welcome: a.sanchez-graells@bristol.ac.uk. The abstract is as follows:

In this paper, I explore the notion of public interest embedded in the Procurement Act 2023. I use this new piece of legislation as a contemporary example of the difficulty in designing a 'public interest centred' system of public procurement regulation. I show how a mix of explicit, referential, and implicit public interest objectives results in a situation where there are multiple sources of objectives contracting authorities need to consider in their decision-making, but there is no prioritisation of sources or objectives. I also show that, despite this proliferation of sources and objectives and due to the unavailability of effective means of judicial challenge or administrative oversight, contracting authorities retain almost unlimited discretion to shape the public interest and 'what it looks like' in relation to the award of each public contract. I conclude with a reflection the need to reconsider the ways in which public procurement can foster the public interest, in light of its limitations as a regulatory tool.

A blog for a book -- Extended for the rest of the summer

It was great to receive a submission for the ‘a book for a blog’ competition, which you can read here. However, it seems that the time was not right, as I would have hoped for a few more entries. And I still have two copies of the book to give away. I am thus extending the deadline until 13 September 2024, in case anyone feels inspired over the summer break. Same conditions as in the original call apply.

I take this chance to wish everyone a good summer, and a good break over it. See you in September!

*Original call for blogs*

I am increasingly interested in the ways public procurement can contribute to slowing down public sector AI adoption. I think slowing down AI adoption is very important because we need time to develop a better understanding of the technology and its systemic implications, and to put much-needed effective safeguards in place.

These are themes I explore in my recent book Digital Technologies and Public Procurement. Gatekeeping and Experimentation in Digital Public Governance (OUP, 2024). I am very interested in what others think of this — including why some may think it is a bad idea. So I am launching ‘a blog for a book’ competition.

If you have ideas on how to use procurement to slow down public sector AI adoption, and/or why it is a good or a bad idea, please send me a blog of up to 2,000 words by 30 June 2024 at a.sanchez-graells@bristol.ac.uk. The top 3 blogs will receive a copy of the book. All interesting blogs will be published (with permission, of course). I look forward to reading your ideas!

Slowing Down Public Sector AI Adoption -- guest post by Saema Jaffer

Amritha R Warrier & AI4Media / Better Images of AI / error cannot generate / CC-BY 4.0.

This guest post by Saema Jaffer* explores the policy approaches that can be followed in the UK, and elsewhere, in relation to the use of AI in the writing and evaluation of tender documents.

It was submitted to the ‘a book for a blog’ competition and has won Saema a copy that will soon be on the post.

Please note Saema’s disclaimer that “I am commercial practitioner and I am writing in a personal capacity. The views expressed in this blog are my own, and do not represent those of my employer”.

Slowing Down Public Sector AI adoption

PROCUREMENT CONFERENCE DAY

I am at an industry conference day, there are rows upon rows of supplier stands, and tables covered with branded pens and tote bags. I am reading a leaflet for a contract management portal when I overhear a conversation between a buyer and a supplier.

“I suppose at this rate, you could just get AI to write all your bids for you?” says a buyer. This elicited a defensive reply from the supplier: “and you could write the tender documents and evaluate all of the bids using AI without ever reading them!”

AI EVALUATIONS

I am absolutely eavesdropping at this point and think: “now there’s a terrible idea!” AI might be fast, but buyers cannot rely on it to be a subject matter expert. What’s more, quite apart from breaching evaluation requirements, what if the AI tool were to misunderstand a supplier commitment, or miss a crucial element revealing non-compliance with the contract? If a bid is incorrectly evaluated, the buyer can hardly hold the AI response generator accountable.

MUTUAL NIGHTMARE

I suppose the bidder’s response would be that as much as the buyer wants to avoid contracting on scope which does not meet their requirements and still less with a supplier who does not meet the scope, the bidder also does not wish to be tied to contractual commitments as part of their bid that they cannot deliver. Simply relying on AI without any human involvement in either client or supplier side, then, presents major problems for both buyer and supplier.

ONE SIZE FITS NO-ONE

But what if the supplier were to use AI to simply write the initial bid, and then review and edit it to make sure there were no commitments that it could not deliver?

After all, the public sector is keen to simplify procurement for small and medium-sized enterprises. An easy way to help those bidders is to use an automated bid writer to help them write their first draft.

Relying on the same AI tools, as I would imagine most SME bidders would do, poses another risk: that the breadth of ideas and sources of creative solutions shrinks. If AI is to be used, we all do need to be aware of what is being lost and be ready to commit to the one-size-fits-all offer drafted by AI.

FEEDING THE AI MACHINE

So, does this mean that bidders should have no restrictions in writing their bids using AI, provided they commit to delivering whatever they end up submitting in their tender?

Not quite. The reason for this is yet another question: what information did the bidder feed the AI response generator to obtain an answer? Was it simply the evaluation question? Possibly.

Was it full extent of the ITT documents, including the scope and the contract? That seems more likely.

What if that ITT contained commercially sensitive information about the buyer’s organisation? That information would have been pasted into a chatbot and then stored…where, and sent to whom?

All the time and effort the buyer would have spent checking for conflicts of interest in the bidding team, signing non-disclosure agreements, being careful to release documents only to pre-qualified suppliers, would mean nothing if the information can simply be released to an unidentifiable person who can disappear into the ether. It’s the same reason why a buyer might be wary of using AI to write invitations to tender.

POTENTIAL SOLUTIONS

So what are my solutions?

Option 1: Place a ban bidders to commit not to using AI. This seems draconian for the vast majority of tenders. I suspect it would also prove ineffective. It’s not so much using a sledgehammer to crack a nut, but using a cardboard cut-out of a sledgehammer to intimidate the nut into cracking.

Option 2: Using the bidder’s own AI solutions. So far, I’ve considered the use of AI that is owned and managed by a third party outside the buyer and bidder relationship. But what if an AI tool is owned by the bidder so that it can control where the information goes?

It’s a partial solution but it does not come without risks. For example, where ethical walls need to be set up within the bidding organisation to prevent conflicts of interest, how can I be sure that a shared AI bot won’t take information from one side of the ethical wall and feed it to the other?

If the bidder is eliminated, can commercially sensitive information that has already been inputted into the system really be destroyed? Once it has been fed into an AI system, can anything really be destroyed?

Option 3: Slowing down AI adoption by establishing a framework with clear guidelines on where AI can be used, and the risks (and consequences) of improper use.

This option recognises that an all-out ban on AI is unlikely to work. Therefore, we need to work on developing suitable criteria for where it can and cannot be used, making clear the inherent risks.

Buyers and suppliers both need to be aware of the risks of using AI as part of tendering processes, and ensure their organisations develop suitable frameworks and ways of working to incorporate AI in a way that manages risks and protects sensitive data.

SLOWING DOWN AI ADOPTION

What requirements should be established to slow down reckless AI adoption and ensure there are effective safeguards in place?

I’ve come up with a shortlist of six practicable ideas below:

1. Contracting authorities should be clear on the scope of AI Use. Specific areas or aspects of the tender submission process where AI cannot be used should be marked out, alongside information which cannot be inputted into third party AI systems.

2. AI systems should include mechanisms for human oversight and intervention, ensuring that critical decisions are reviewed and approved by humans. Linked to this, bid writers must have been trained in how to effectively and ethically use AI technologies before participating in a tendering process.

3. The bidding organisation must confirm, as part of their offer, that everything that the AI tool has produced has been separately verified by the bidding entity and that the AI-drafted submission constitutes a firm offer from the bidder.

4. Bidders must explicitly disclose if and how AI technologies were used in the preparation of their tender submissions. This must be auditable, allowing the contracting authority to verify the integrity and reliability of the AI tools used.

5. Bidders should provide clear explanations of AI generated outputs, making it understandable how conclusions or recommendations were reached.

6. Independent third-party audits or certifications must be required to validate the AI technology's adherence to specified standards in relation to:

• Adherence to Technical Standards, so that AI systems are quality assured.

• Adherence to ethical guidelines, ensuring that there is no theft of intellectual property.

• Compliance with data protection laws, ensuring that any personal data used or processed is secure and handled in accordance with legal requirements.

• Non-discrimination, because the AI tools must not introduce bias or discrimination into the tendering process. Bidders should provide evidence of measures taken to prevent such issues.

The adoption of such a framework should give confidence that AI technologies are used responsibly and ethically in the procurement process.

In the words of Alan Turing, the creator of modern computing: “We can only see a short distance ahead, but we can see plenty that needs to be done.”

SAEMA JAFFER

Saema Jaffer is a procurement and commercial strategy professional with experience across the public sector. She has a strong foundation in managing the full procurement lifecycle, ensuring compliance with regulations and facilitating cultural change through strategic process improvements. Saema Jaffer is interested in changes brought by the Procurement Act, artificial intelligence and innovation within the construction sector. You can connect with her on LinkedIn.

Creating (positive) friction in AI procurement

I had the opportunity to participate in the Inaugural AI Commercial Lifecycle and Procurement Summit 2024 hosted by Curshaw. This was a very interesting ‘unconference’ where participants offered to lead sessions on topics they wanted to talk about. I led a session on ‘Creating friction in AI procurement’.

This was clearly a counterintuitive way of thinking about AI and procurement, given that the ‘big promise’ of AI is that it will reduce friction (eg through automation, and/or delegation of ‘non-value-added’ tasks). Why would I want to create friction in this context?

The first clarification I was thus asked for was whether this was about ‘good friction’ (as opposed to old bad ‘red tape’ kind of friction), which of course it was (?!), and the second, what do I mean by friction.

My recent research on AI procurement (eg here and here for the book-long treatment) has led me to conclude that we need to slow down the process of public sector AI adoption and to create mechanisms that bring back to the table the ‘non-AI’ option and several ‘stop project’ or ‘deal breaker’ trumps to push back against the tidal wave of unavoidability that seems to dominate all discussions on public sector digitalisation. My preferred solution is to do so through a system of permissioning or licencing administered by an independent authority—but I am aware and willing to concede that there is no political will for it. I thus started thinking about second-best approaches to slowing public sector AI procurement. This is how I got to the idea of friction.

By creating friction, I mean the need for a structured decision-making process that allows for collective deliberation within and around the adopting institution, and which is supported by rigorous impact assessments that tease out second and third order implications from AI adoption, as well as thoroughly interrogating first order issues around data quality and governance, technological governance and organisational capability, in particular around risk management and mitigation. This is complementary—but hopefully goes beyond—emerging frameworks to determine organisational ‘risk appetite’ for AI procurement, such as that developed by the AI Procurement Lab and the Centre for Inclusive Change.

The conversations the focus on ‘good friction’ moved in different directions, but there are some takeaways and ideas that stuck with me (or I managed to jot down in my notes while chatting to others), such as (in no particular order of importance or potential):

• the potential for ‘AI minimisation’ or ‘non-AI equivalence’ to test the need for (specific) AI solutions—if you can sufficiently approximate, or replicate, the same functional outcome without AI, or with a simpler type of AI, why not do it that way?;

• the need for a structured catalogue of solutions (and components of solutions) that are already available (sometimes in open access, where there is lots of duplication) to inform such considerations;

• the importance of asking whether procuring AI is driven by considerations such as availability of funding (is this funded if done with AI but not funded, or hard to fund at the same level, if done in other ways?), which can clearly skew decision-making—the importance of considering the effects of ‘digital industrial policy’ on decision-making;

• the power (and relevance) of the deceptively simple question ‘is there an interdisciplinary team to be dedicated to this, and exclusively to this’?;

• the importance of knowledge and understanding of the tech and its implications from the beginning, and of expertise in the translation of technical and governance requirements into procurement requirements, to avoid ‘games of chance’ whereby the use of ‘trendy terms’ (such as ‘agile’ or ‘responsible’) may or may not lead to the award of the contract to the best-placed and best-fitting (tech) provider;

• the possibility to adapt civic monitoring or social witnessing mechanisms used in other contexts, such as large infrastructure projects, to be embedded in contract performance and auditing phases;

• the importance of understanding displacement effects and whether deploying a solution (AI or automation, or similar) to deal with a bottleneck will simply displace the issue to another (new) bottleneck somewhere along the process;

• the importance of understanding the broader organisational changes required to capture the hoped for (productivity) gains arising from the tech deployment;

• the importance of carefully considering and resourcing the much needed engagement of the ‘intelligent person’ that needs to check the design and outputs of the AI, including frontline workers and those at the receiving end of the relevant decisions or processes and the affected communities—the importance of creating meaningful and effective deliberative engagement mechanisms;

• relatedly, the need to ensure organisational engagement and alignment at every level and every step of the AI (pre)procurement process (on which I would recommend reading this recent piece by Kawakami and colleagues);

• the need to assess the impacts of changes in scale, complexity, and error exposure;

• the need to create adequate circuit-breakers throughout the process.

Certainly lots to reflect on and try to embed in future research and outreach efforts. Thanks to all those who participated in the conversation, and to those interested in joining it. A structured way to do so is through this LinkedIn group.

Not my standard -- presentation at GR'24

I am attending Global Revolution 24 at the University of Nottingham. This is always an interesting conference and a good platform to catch up with colleagues and friends from all over the world. I spent Day 1 attending the four workshops of the AI and procurement stream that Jessica Tillipman (GW Law) and I co-organised. This was a day full of insightful and engaging discussions, and I took a lot of ideas for further research and policy work, including on the need for a different approach to conflicts of interest that can be embedded in the tech (more on that some other time).

My presentation featured a unicorn metaphor to synthesise my views that the current approach to the regulation of public sector AI adoption through contract and, in particular, through the EU AI Act and the pilot model AI contractual clauses is bound to fail and requires a rethink. Like ‘haunting for a unicorn’, pursuing ‘trustworthy AI’ is a major challenge because we are after something that we have never seen and can only describe on the basis of our imagination. The absence of clear metrics to express open-ended concepts, such as fairness, and the impossibility of designing effective contract-based governance for the socio-technical embedding of the AI, inevitably create excessive indefinition in the regulatory approach and leave it open to commercial determination and regulatory tunnelling.

Moreover, the ‘regulation by contract’ strategy based on a skeletal blueprint (the model AI clauses) that requires significant amounts of technical and commercial expertise for public buyers to fill in is like asking a child to copy an outline of a unicorn and colour it in. The outcome may be endearing or even artsy. But it may hardly resemble a unicorn, which we may only see by stretching our imagination. And deploying ‘AI on AI’ will not solve the situation because ‘technical fixes’ cannot make up for the indeterminacy of the regulatory data and goals—much like an automated description of the unicorns drawn by kids (below) only approximated the target by identifying on picture as that of a horse, and the other as that of a child!

I am not sure if the metaphor totally landed, but I enjoyed the presentation and there was interesting discussion in the Q&A afterwards. The presentation is available here and I would be more than happy to continue the conversation.

'Pro bono' or 'land and expand'? -- problematic 'zero-value' or 'free' contracts for digital innovation

Max Gruber / Better Images of AI / Banana / Plant / Flask / CC-BY 4.0.

The UK’s Ministry of Justice recently held an 8-day competition to select a software and programming consultancy to carry out a ‘pro-bono proof-of-concept process to explore methods for human-in-the-loop triage through GenAI’ (let’s not get bogged down on the technical details…).

This opportunity was not advertised under public procurement rules because the Ministry of Justice estimated the value of the contract at £0, as there would be no (direct, monetary) payments to the consultancy; it was clear that this would be ‘a pro bono contract. The [proof-of-concept] is expected to be completed in a 6 to 8 week period.’

Although the notice made it clear that the Ministry of Justice does ‘not anticipate that this project will be followed by further procurement on future development using the AI human in the loop triage’, because the main purpose is to ‘help accelerate existing work, test and prove different solutions/tools, share learning to help inform future work and designs, and transfer skills back into the [in-house] team throughout’—this approach seems problematic, in at least two ways.

First, it raises questions on whether, even as a ‘non-procurement’ opportunity, this was carried out in a proper way aligned with best practice. An 8-day window to express interest seems very short, especially as potentially interested consultants/consultancies were given very limited information to estimate the scope of works and understand the cost (to them) of supporting the development of the proof-of-concept on a pro bono basis.

At the same time, meeting the ‘minimalistic’ selection criteria and technical requirements could have been tricky, especially in such short time scale, as they contained some aspects that would have been difficult or impossible to meet other than by entities that already met the requirements at the time of the notice (eg obtaining BPSS security clearance is presumably not instantaneous) and were very familiar with the open source and other design standards referred to.

It does not seem unreasonable to suspect that the Ministry of Justice may have been in prior talks with some company/ies potentially interested in providing those pro bono services and that this created an insider advantage in making sense of the otherwise seemingly insufficient details given in the notice. This would make a mockery of the advertisement of the opportunity, even if not under procurement rules.

Cynically, it is also not clear what would happen if the current view changed and the Ministry later decided to procure the further stages of development — for seeking external input already at proof-of-concept stage (strongly) suggests that the in-house teams do not have (confidence in their) digital skills as required to carry out such digital innovation work.

In that case, an argument could be made that only the consultancy that participated in the proof-of-concept could provide the services for some specific technical / know how reason and that could seek to be used as justification for a direct award. This is a non-trivial risk. Even if this was not the case, a further procurement would create significant issues of prior involvement and would require deploying complicated solutions to try to neutralise the advantage gained by the consultancy. Overall, this does not seem like a fool-proof way of managing early collaborations in digital innovation projects.

Second, and perhaps more controversially, it is also not entirely clear to me that the contract would actually have ‘zero-value’ for the consultancy and was thus not really covered by the procurement rules.

The notice makes it clear that the Ministry of Justice would seek to retain the knowledge arising from the pilot by sharing it ‘within internal teams who can carry on building on top of the deliverables, but also with the wider teams that that are interested in using these new tools.’ However, this does not mean that the consultancy will not acquire potentially (almost) exclusive rights over the knowledge vis-à-vis third parties. There is no indication that the Ministry of Justice will publish the outcome of the project and, consequently, there seems to be no obstacle to the contribution of the Ministry being appropriated and incorporated into the consultancy’s know how (or future IP).

On close analysis of the terms of the ‘opportunity’ advertised by the Ministry of Justice, it is also clear that the Ministry would make available ‘A limited number of redacted example cases [to] be used for experimentation’ and that the consultant would be given access to the Ministry’s then current thinking and expertise on the project. These can also be valuable non-monetary assets on their own.

All in all, this is in itself worrying and raises questions on whether this was actually a £0 contract (from the perspective of the consultancy). Two legal aspects are relevant here. First, (partly pre-Brexit) ECJ case law distinguishes between pecuniary interest and monetary payments (IBA Molecular Italy and Tax-Fin-Lex), and stresses that ‘It is clear from the usual legal meaning of “for pecuniary interest” that those terms designate a contract by which each of the parties undertakes to provide a service in exchange for another’ (IBA Molecular Italy, C‑606/17, EU:C:2018:843, para 28). Whether this case—through the mix of access to anonymised examples, access to in-house know-how and possibility to retain know-how over the proof-of-concept and related learning vis-à-vis third parties—crosses the threshold and would have required advertising as a ‘procurement’ opportunity is at least arguable.

Second, this was clearly always going to be a ‘loss’ contract for the ‘pro bono volunteer’, and the fact that it is advertised as ‘pro bono’ solely masks the fact that the Ministry (a contracting authority in general terms) imposed on tenderers a requirement to submit abnormally low tenders. This raises questions on ‘mixed pro bono’ justifications that could be put forward in other cases to justify that a very low-cost tender is not actually abnormally low, but in the context of a procurement estimated at a value above the relevant threshold. Such justifications would in principle seem impermissible and, more generally, there are very big questions as to why the public sector should be seeking to obtain ‘free consultancy’—other than through fully open mechanisms, such as eg hackatons (is that still a thing) or other sorts of non-procurement competitions.

Which leads to the related question why would anyone willingly enter into such a contract? At least a partial explanation is that the ‘pro bono volunteer’ would most likely see a possible market advantage—for there are endless possibilities to carry out pro bono work otherwise. If not in this case, in most cases where similar conditions arise, there is a clear ‘marketing’ drive that can be masked as pro bono generosity. This can easily become an embedded element of strategies to ‘land and conquer’, or to build a portfolio of pro bono projects that is later used eg to demonstrate technical capability for qualitative selection purposes in future procurement processes (with other contracting authorities).

Overall, I think this is an example of worrying trends in the (side-stepping of) ‘procurement’ of digital innovation support services, and that such trends should be resisted. Only proper procurement processes and robust guardrails to safeguard from the risks of capture, competitive distortion and more generally long-term difficulties in ensuring competitive tension for digital innovation contracts, can minimise the consequences of arrangements that seem too good to be true.

A blog for a book -- How to use procurement to slow down public sector AI adoption? And should we (not)?

I am increasingly interested in the ways public procurement can contribute to slowing down public sector AI adoption. I think slowing down AI adoption is very important because we need time to develop a better understanding of the technology and its systemic implications, and to put much-needed effective safeguards in place.

These are themes I explore in my recent book Digital Technologies and Public Procurement. Gatekeeping and Experimentation in Digital Public Governance (OUP, 2024). I am very interested in what others think of this — including why some may think it is a bad idea. So I am launching ‘a blog for a book’ competition.

If you have ideas on how to use procurement to slow down public sector AI adoption, and/or why it is a good or a bad idea, please send me a blog of up to 2,000 words by 30 June 2024 at a.sanchez-graells@bristol.ac.uk. The top 3 blogs will receive a copy of the book. All interesting blogs will be published (with permission, of course). I look forward to reading your ideas!

Meaning, AI, and procurement -- some thoughts

©Ausrine Kuze, Distorted Reality, 2021.

James McKinney and Volodymyr Tarnay of the Open Contracting Partnership have published ‘A gentle introduction to applying AI in procurement’. It is a very accessible and helpful primer on some of the most salient issues to be considered when exploring the possibility of using AI to extract insights from procurement big data.

The OCP introduction to AI in procurement provides helpful pointers in relation to task identification, method, input, and model selection. I would add that an initial exploration of the possibility to deploy AI also (and perhaps first and foremost) requires careful consideration of the level of precision and the type (and size) of errors that can be tolerated in the specific task, and ways to test and measure it.

One of the crucial and perhaps more difficult to understand issues covered by the introduction is how AI seeks to capture ‘meaning’ in order to extract insights from big data. This is also a controversial issue that keeps coming up in procurement data analysis contexts, and one that triggered some heated debate at the Public Procurement Data Superpowers Conference last week—where, in my view, companies selling procurement insight services were peddling hyped claims (see session on ‘Transparency in public procurement - Data readability’).

In this post, I venture some thoughts on meaning, AI, and public procurement big data. As always, I am very interested in feedback and opportunities for further discussion.

Meaning

Of course, the concept of meaning is complex and open to philosophical, linguistic, and other interpretations. Here I take a relatively pedestrian and pragmatic approach and, following the Cambridge dictionary, consider two ways in which ‘meaning’ is understood in plain English: ‘the meaning of something is what it expresses or represents’, and meaning as ‘importance or value’.

To put it simply, I will argue that AI cannot capture meaning proper. It can carry complex analysis of ‘content in context’, but we should not equate that with meaning. This will be important later on.

AI, meaning, embeddings, and ‘content in context’

The OCP introduction helpfully addresses this issue in relation to an example of ‘sentence similarity’, where the researchers are looking for phrases that are alike in tender notices and predefined green criteria, and therefore want to use AI to compare sentences and assign them a similarity score. Intuitively, ‘meaning’ would be important to the comparison.

The OCP introduction explains that:

Computers don’t understand human language. They need to operate on numbers. We can represent text and other information as numerical values with vector embeddings. A vector is a list of numbers that, in the context of AI, helps us express the meaning of information and its relationship to other information.

Text can be converted into vectors using a model. [A sentence transformer model] converts a sentence into a vector of 384 numbers. For example, the sentence “don’t panic and always carry a towel” becomes the numbers 0.425…, 0.385…, 0.072…, and so on.

These numbers represent the meaning of the sentence.

Let’s compare this sentence to another: “keep calm and never forget your towel” which has the vector (0.434…, 0.264…, 0.123…, …).

One way to determine their similarity score is to use cosine similarity to calculate the distance between the vectors of the two sentences. Put simply, the closer the vectors are, the more alike the sentences are. The result of this calculation will always be a number from -1 (the sentences have opposite meanings) to 1 (same meaning). You could also calculate this using other trigonometric measures such as Euclidean distance.

For our two sentences above, performing this mathematical operation returns a similarity score of 0.869.

Now let’s consider the sentence “do you like cheese?” which has the vector (-0.167…, -0.557…, 0.066…, …). It returns a similarity score of 0.199. Hooray! The computer is correct!

But, this method is not fool-proof. Let’s try another: “do panic and never bring a towel” (0.589…, 0.255…, 0.0884…, …). The similarity score is 0.857. The score is high, because the words are similar… but the logic is opposite!

I think there are two important observations in relation to the use of meaning here (highlighted above).

First, meaning can hardly be captured where sentences with opposite logic are considered very similar. This is because the method described above (vector embedding) does not capture meaning. It captures content (words) in context (around other words).

Second, it is not possible to fully express in numbers what text expresses or represents, or its importance or value. What the vectors capture is the representation or expression of such meaning, the representation of its value and importance through the use of those specific words in the particular order in which they are expresssed. The string of numbers is thus a second-degree representation of the meaning intended by the words; it is a numerical representation of the word representation, not a numerical representation of the meaning.

Unavoidably, there is plenty scope for loss, alteration or even inversion of meaning when it goes through multiple imperfect processes of representation. This means that the more open textured the expression in words and the less contextualised in its presentation, the more difficult it is to achieve good results.

It is important to bear in mind that the current techniques based on this or similar methods, such as those based on large language models, clearly fail on crucial aspects such as their factuality—which ultimately requires checking whether something with a given meaning is true or false.

This is a burgeoning area of technnical research but it seems that even the most accurate models tend to hover around 70% accuracy, save in highly contextual non-ambiguous contexts (see eg D Quelle and A Bovet, ‘The perils and promises of fact-checking with large language models’ (2024) 7 Front. Artif. Intell., Sec. Natural Language Processing). While this is an impressive feature of these tools, it can hardly be acceptable to extrapolate that these tools can be deployed for tasks that require precision and factuality.

Procurement big data and ‘content and context’

In some senses, the application of AI to extract insights from procurement big data is well suited to the fact that, by and large, existing procurement data is very precisely contextualised and increasingly concerns structured content—that is, that most of the procurement data that is (increasingly) available is captured in structured notices and tends to have a narrowly defined and highly contextual purpose.

From that perspective, there is potential to look for implementations of advanced comparisons of ‘content in context’. But this will most likely have a hard boundary where ‘meaning’ needs to be interpreted or analysed, as AI cannot perform that task. At most, it can help gather the information, but it cannot analyse it because it cannot ‘understand’ it.

Policy implications

In my view, the above shows that the possibility of using AI to extract insights from procurement big data needs to be approched with caution. For tasks where a ‘broad brush’ approach will do, these can be helpful tools. They can help mitigate the informational deficit procurement policy and practice tend to encounter. As put in the conference last week, these tools can help get a sense of broad trends or directions, and can thus inform policy and decision-making only in that regard and to that extent. Conversely, AI cannot be used in contexts where precision is important and where errors would affect important rights or interests.

This is important, for example, in relation to the fascination that AI ‘business insights’ seems to be triggering amongst public buyers. One of the issues that kept coming up concerns why contracting authorities cannot benefit from the same advances that are touted as being offered to (private) tenderers. The case at hand was that of identifying ‘business opportunities’.

A number of companies are using AI to support searches for contract notices to highlight potentially interesting tenders to their clients. They offer services such as ‘tender summaries’, whereby the AI creates a one-line summary on the basis of a contract notice or a tender description, and this summary can be automatically translated (eg into English). They also offer search services based on ‘capturing meaning’ from a company’s website and matching it to potentially interesting tender opportunities.

All these services, however, are at bottom a sophisticated comparison of content in context, not of meaning. And these are deployed to go from more to less information (summaries), which can reduce problems with factuality and precision except in extreme cases, and in a setting where getting it wrong has only a marginal cost (ie the company will set aside the non-interesting tender and move on). This is also an area where expectations can be managed and where results well below 100% accuracy can be interesting and have value.

The opposite does not apply from the perspective of the public buyer. For example, a summary of a tender is unlikely to have much value as, with all likelihood, the summary will simply confirm that the tender matches the advertised object of the contract (which has no value, differently from a summary suggesting a tender matches the business activities of an economic operator). Moreover, factuality is extremely important and only 100% accuracy will do in a context where decision-making is subject to good administration guarantees.

Therefore, we need to be very careful about how we think about using AI to extract insights from procurement (big) data and, as the OCP introduction highlights, one of the most important things is to clearly define the task for which AI would be used. In my view, there are much more limited tasks than one could dream up if we let our collective imagination run high on hype.

Interesting AG Opinion on State aid analysis of procurement compliance, definition of public works contracts, and ‘strategic’ use of remedies by contracting authorities (C-28/23)

On 11 April 2024, AG Campos Sánchez-Bordona delivered his Opinion in NFŠ (C-28/23, EU:C:2024:306). The NFŠ Opinion is very interesting in three respects. First, in addressing some aspects of the definition of public works contracts that keep coming up in litigation in relation to relatively complex real estate transactions. Second, in addressing the effects of a State aid decision on the assessment of compliance with procurement law of the legal structure used to implement the aid package. Third, in addressing some limits on the ‘strategic’ use of remedies by contracting authorities that have breached procurement law. Before providing some comments on the Opinion, I need to make two disclaimers.

The first one is that, exceptionally, I have been involved in the legal proceedings before the ECJ. At the request of NFŠ, I wrote an expert statement addressing some of the issues raised by the case. I am very pleased to see that my own legal analysis coincides with that of AG Campos Sánchez-Bordona, and I hope the Court will also share it in the forthcoming Judgment.

Second, it is worth stressing that this is not a bread and butter procurement case and referring to the legal structure can be cumbersome or confusing if not done precisely. Unfortunately, this has happened in the English translation of the AG Opinion, which is rather poor in some areas. In particular, crucial paragraphs 81 and 96 are incorrectly translated and convey a confusing position. To avoid those issues, I rely on my own translation of the Spanish and French versions of the Opinion (and highlight it where my own translation deviates from the ECJ’s one by placing the relevant parts in [square brackets and italics]).

Background

In short, the case arises from a dispute between the Slovak Government and NFŠ in relation to the Slovak national football stadium. Despite having provided State aid for the construction of the stadium, the State is now unwilling to purchase it from NFŠ in the terms of the aid package. This has resulted in domestic litigation. The request from a preliminary reference emerges in this context.

In 2013, the Slovak Government entered into a grant agreement with NFŠ to support the construction of the national football stadium in Bratislava. However, construction did not immediately proceed and the level of financial support was in need of review. The grant agreement was revised in 2016 (the ‘grant agreement’). In addition to the grant for the construction of the stadium, the Slovak Government also granted NFŠ a unilateral put option to sell the stadium to the State, under certain conditions, during the five years following its completion (the ‘agreement to enter into a future sales agreement’ or ‘AFSA’).

Slovakia notified this set of agreements to the European Commission as State aid. In 2017, the Commission declared those measures to be compatible with the internal market by Decision State Aid SA.46530. The State aid Decision made it clear that the total volume of aid comprised the direct grant plus the value of the put option, and that those modalities and that level of aid were justified in view of the need to provide sufficient financial incentives to get the stadium developed. In relation to the put option, the Commission stated that ‘The option given to the beneficiary allows it to sell the Stadium back to the State in case it wishes to do so. Should the beneficiary decide to exercise the option, the Stadium would become a property of the State’ (para 22). In relation to the obligation to subject the construction of the stadium to competitive public procurement, the State aid Decision also explicitly stated that ‘The construction works financed through the grant … will be subject to a competitive process, respecting the applicable procurement rules’ (para 8).

NFŠ undertook the development of the stadium and awarded contracts for different parts of the works under competitive tender procedures compliant with the Slovak transposition of EU law. All tenders were advertised in the Official Journal of the European Union and in the Slovak official journal. Once the stadium was completed and in operation, NFŠ decided to exercise the put option and called on the Slovak Government to purchase the stadium in the terms foreseen in AFSA.

Simply put, in order to try to avoid the obligation to purchase the football stadium in the terms set out in AFSA, the Slovak Government is arguing that the agreements are null and void because, combined and from the outset, the grant agreement and AFSA would have had the unavoidable effect of getting the stadium built and transferred to the State, and thus cover up the illegal direct award of a public works contract to NFŠ. This part of the dispute concerns the definition of ‘public works contracts’ under Directive 2014/24/EU (section 1 below).

Relatedly, the Slovak Government states that despite containing explicit references to the tendering of the construction of the stadium, the State aid Decision cannot preempt a fresh assessment of the compliance of this legal structure with EU procurement rules. Perhaps surprisingly, this position has been supported by the European Commission, which denied that the explicit mention of compliance with procurement law formed an integral part of its assessment of the compatibility of the set of agreements with EU internal market law. This is a crucial issue and the outcome of this case can provide much needed clarity on the extent to which the Commission does, and indeed must, take procurement law into account in the assessment of State aid measures that involve the award of public contracts. This part of the dispute thus concerns the effect of State aid decisions relating to aid packages with a procurement element (section 2 below).

Finally, it is also important in the case that the State seeks confirmation of the possibility of having the ineffectiveness of the grant agreement and AFSA recognised ex tunc under domestic law, without this being a breach of the Remedies Directive. This relates to the ‘strategic’ use of procurement remedies by contracting authorities that have breached procurement law (section 3 below).

The AG Opinion deals with these issues and is interesting in all respects, but specially the latter two, where it breaks new ground.

1. Definition of a ‘public works contract’

The first issue addressed in the AG Opinion concerns whether the grant agreement and AFSA create such a set of obligations on NFŠ as beneficiary of the aid and developer of the stadium that, in reality, they amount to the illegal direct award of a public works contract for the construction of the stadium. There are three main issues that require detailed consideration:

• whether the contractor had assumed a legally enforceable direct or indirect obligation to carry out the works;

• whether the works should be executed in accordance with the requirements specified by the contracting authority, which thus had decisive influence over the project; and

• whether the contracting authority would obtain a direct economic benefit.

The AG Opinion provides a helpful summary of the case law on these issues (see paras 52-54) and additional guidance on how to apply them in the case, raising significant questions on whether these criteria were met—although the final assessment must be carried out by the referring court.

Legally Enforceable Obligation

First, the Opinion stresses that it is unclear that NFŠ was placed under a legally enforceable obligation to build and transfer the stadium as a result of the grant agreement and AFSA. Importantly, the AG distinguishes the existence of an enforceable obligation to carry out the works from the existence of legal consequences from deciding not to do so. As the Opinion makes clear, the simple existence of the agreements to subsidise the development of the stadium does not ‘support the inference that the Slovak State would have any right to take legal action against NFŠ to compel it to build the stadium should that undertaking ultimately decide not to do so. [A different issue is whether], in that event, NFŠ would not have received the grant, or would have lost it, or would [have been] obliged to pay it back. This in itself, however, has nothing to do with the performance of a works contract’ (para 59).

This is important because it sets the threshold at which a ‘commitment’ to carry out works becomes a legally enforceable obligation for the purposes of EU public procurement law. It reflects an understanding that there has to be a right (in principle) to require specific performance (performance in natura), not solely the existence of legal consequences arising from a decision not to follow through with such a commitment. This is further supported in the fact that ‘the mere grant of a State subsidy involving the [disbursement] of public funds (in the present case, for the purpose of constructing a stadium) does not in itself amount to the conclusion of a public works contract. As recital 4 of Directive 2014/24 states, “the Union rules on public procurement are not intended to cover all forms of disbursement of public funds, but only those aimed at the acquisition of works, supplies or services for consideration by means of a public contract”’ (para 48, underline emphasis in the original).

The Opinion further stresses that:

in order for there to be a genuine works contract, it is essential that the successful tenderer should specifically take on the obligation to carry out the works forming the subject of the acquisition and that that obligation should be legally enforceable [in court]. The contracting authority … must acquire the [building] on which the works are carried out and, [where applicable], [be able to] take legal action [in court] to compel the tenderer awarded the contract to [transfer it], if it holds [legal title covering the encumbrance of the works for the purposes of public use] (para 60, underline emphasis in the original).

This concerns the legal enforceability of the put option from the perspective of the State. In that regard, it will be necessary for the referring court to establish ‘whether NFŠ, once the sports infrastructure had been built, had a legally enforceable obligation to transfer it to the Slovak State, which the latter could assert’ (para 61). The Opinion suggests that this is highly implausible, given that ‘all the indications are that the agreement to enter into a future sales agreement gave NFŠ the option either to remain the owner of the stadium and continue to operate it (or assign its operation to third parties), or to transfer it [to] the Slovak State, if [doing so suited that undertaking]’ (para 62).

Moreover, and this is a crucially interesting aspect of the case, the Opinion stresses that the assessment of the legal enforceability of the put option had already been the object of analysis by the European Commission in its State aid decision and that the Commission had confirmed that it enabled NFŠ ‘ (but does not oblige it) to sell the infrastructure to the Slovak State if that undertaking wishes to do so’ (para 63). This will be particularly relevant in view of the effects of the State aid Decision discussed in section 2 below.

Specifications by the Contracting Authority

A second issue of relevance in the case is that the aid package required for the stadium to meet ‘UEFA Regulations on the construction of category 4 stadiums and those contained in the general Slovak rules on sports infrastructure projects’ (para 65). This raises the question whether the contracting authority could exercise ‘decisive influence over the construction project’ by requiring compliance with those requirements (ibid) and participating in a monitoring committee. The Opinion focuses on the material impact of those circumstances on the development of the project.

Interestingly, the Opinion stresses that ‘UEFA criteria … consist of a number of mandatory parameters in relation to the minimum structural requirements which a stadium must meet in order to be classified in a certain category. However, those criteria are amenable to a variety of architectural solutions that can be developed within very broad margins of professional creativity’; and that ‘The design of football stadiums that comply with the UEFA criteria allows for an extensive range of creative alternatives, both in the external configuration of the stadium and in the structuring of its internal amenities. Those criteria do not … contain the detailed technical solutions which a true proprietor of the work could impose on the tenderer awarded the contract’ (paras 66-67, reference omitted).

This part of the Opinion is interesting in the context of drawing the boundaries between real estate transactions that will be caught or not by the procurement rules because it comes to develop the guidance offered by previous case law (recently C‑537/19, EU:C:2021:319) on the extent to which the specifications need to be sufficiently detailed to exceed the usual requirements of a tenant (C‑536/07, EU:C:2009:664). The further clarification is, in my view, that the specifications should be such as to significantly constrain or predetermine architectural solutions in the design of the works.

Direct Economic Benefit to the Contracting Authority

The Opinion directly refers to the case law on the need that ‘In a public works contract, the contracting authority receives a service consisting of the realisation of works which it seeks to obtain and which has a direct economic benefit for it’ (para 52). In that regard, the Opinion stresses that it is not sufficient for the Slovak State to have an ‘interest (and subsequent indirect benefit) … confined to the generic promotion of the national sport’ (para 64). This is also important because it clarifies the threshold of ‘directness’ and magnitude of the interest that must arise for a legal transaction to be classed as a public contract.

2. Effect of State aid decisions relating to aid packages with a procurement element

Perhaps the most interesting issue that the AG Opinion deals with is the extent to which a State aid Decision declaring a legal structure with explicit procurement implications compatible with the internal market pre-empts a separate assessment of its compliance with EU public procurement law.

As mentioned above, in the NFŠ case, the State aid notification had provided details on the grant agreement and AFSA, and made it explicit that the beneficiary of the aid would run public tenders for the competitive award of contracts for the subsidized works. The Commission explicitly referred to this in the Decision, indicating that ‘The construction works financed through the grant … will be subject to a competitive process, respecting the applicable procurement rules’.

As a starting point, AG Campos stresses that, consequently, any assessment of compliance with EU law cannot ignore ‘the considerations set out by the Commission in Decision SA.46530 in connection with the content of the grant agreement and the agreement to enter into a future sales agreement, where it found that, through those agreements, the Slovak State had granted public aid compatible with the internal market’ (para 47).

In more detail, the AG stresses that ‘the Commission examined the grant agreement and the agreement to enter into a future sales agreement. In Decision SA.46530, it evaluated the public aid associated with those agreements and declared it to be compatible with the internal market’ and that ‘A reading of paragraph 8 shows that what mattered to the Commission was that the construction of the stadium (which represents the very essence of a works contract, whether public or private) should be [subjected to] a competitive process respecting the rules applicable to public contracts’ (para 72, reference omitted, and para 74, underline emphasis in the original).

The AG stressed that the Commission confirmed that this was ‘an essential condition for the compatibility of the aid with the internal market’ (para 73). This led the AG to find that the State aid Decision had the effect of triggering the application of the EU procurement rules by NFŠ, ‘which was put in a situation analogous to that of a contracting authority’ (para 75) and, implicitly, that compliance with EU procurement law concerned the contract/s for the works to be tendered by NFŠ, not the award of the State aid to NFŠ.

Crucially, AG Campos spelled out the implications of such consideration by the Commission of the procurement implications of the State aid package within the procedure for State aid control. In his view:

The Commission can actively intervene in defence of competition where public procurement does not comply with the rules laid down in, inter alia, Directive 2014/24 in order to safeguard this objective [to ensure that “public procurement is opened up to competition”]. I do not see any reason why it should not do so when faced with an examination of the viability of State aid measures resulting from agreements concluded by public authorities with private entities.

In particular, it is my view that the Commission could not have failed to examine whether the form in which the public aid granted to NFŠ was structured masked the existence of a public contract which should have been put out to tender. To my mind, it did so implicitly, which explains paragraph 8 of its Decision SA.46530.

In short, Decision SA.46530 is based on the premiss that there was no obligation to transfer ownership of the stadium to the Slovak Republic. That assumption, to which I have already referred, cannot be called into question by the referring court, which must respect the Commission’s assessment of the factors determining the existence of State aid (paras 77-79, underline emphasis in the original, other emphasis added).

This sets out two important implications. The first one, of relatively more limited scope but crucial practical importance, is that as an implicit effect of the Commission’s monopoly of enforcement of the State aid rules, a previous State aid decision does preclude a fresh assessment of a legal structure for the purposes of its compliance with public procurement law. A national court called upon to assess such legal structure cannot call the Commission’s assessment and must respect the Commission’s assessment of the factors determining the existence of State aid. In the NFŠ case, given that the Commission had clearly assessed the put option as entirely discretionary for NFŠ, it is not now possible for the referring court to deviate from that assessment and consider that it established an obligation legally enforceable by the Slovak Government. This carries the additional implication that the legal structure cannot be classed as a public works contract for the purposes of Directive 2014/24/EU.

Therefore, on this point, the AG could have been clearer and made it explicit that, even if the referring court is in principle tasked with the clarification of the relevant circumstances and their legal classification, in this case and given the prior binding assessment of the Commission, it is not possible to rely on the put option under AFSA to class the legal structure as a public works contract because there was no legally binding obligation concerning the transfer of the stadium. However, this conclusion is plain from the joint reading of paras 63 and 79 of the Opinion.

The second implication is that, by way of principle, there is a general obligation for the Commission to assess the compatibility with the EU public procurement law of State aid measures that have procurement implications. I think this is a clarification of the existing case law on the duty on the Commission to assess State aid measures for compliance with other sets of EU internal market law and a very welcome development given the very close connection between State aid and procurement, as evidenced amongst other sources in the Commission’s guidance on the notion of State aid.

3. ‘Strategic’ use of procurement remedies by contracting authorities

A final issue which is also very interesting is that the case provides a very uncommon set of circumstances whereby the same authority that had granted State aid and accepted the legality of the legal structure creating the put option under which it would be purchasing the stadium is later on (under different political circumstances) trying to get out of its obligations and, in doing so, seeks to gain support for its position from the rules on contractual ineffectiveness in the Remedies Directive—with any such effectiveness arising from its own alleged circumvention of EU procurement law.

Of the treatment of this issue in the AG Opinion, I think the following passages are particularly relevant:

Directive 89/665 is not designed to protect the public authorities from infringements which they themselves have committed, but to allow those who have been harmed by the actions of those contracting authorities to challenge them.

Article 2d of Directive 89/665 presupposes that a person entitled to challenge the conduct of the contracting authority has made use of the relevant review procedure. If, at the end of that review, the body adjudicating on it declares the contract in question to be ineffective, the provisions contained in the various paragraphs of that article will be triggered. As I have already said, however, Directive 89/665 does not make provision for the contracting authority to challenge its own decisions.

[A different issue is whether] national law provides ways for a public authority (or an administrative review body) to review the legality of its previous decisions. Such an eventuality is governed not by Directive 89/665 but by the relevant provisions of national law, in accordance with which it will fall to be determined to what extent an exception may be made to the classic rule venire contra factum propium nulli conceditur (paras 88-90, reference omitted, emphasis added).

I think this may have not needed spelling out except in a bizarre case such as NFŠ. However, I also think that this clarification can have broader implications in relation to the (separate) trend to recognize ‘subjective rights’ to contracting authorities under EU public procurement law (see eg in relation to exclusion decisions in (C-66/22, EU:C:2023:1016; for discussion see here).

Final thoughts

I think NFŠ will be an important case and I very much hope that the Court will follow AG Campos in this case. I also hope that the clarification of the aspects concerning the effect of State aid decisions and, more importantly, the general duty for the Commission to assess compliance of State aid measures with EU public procurement law, will explicitly feature in the judgment of the Court. I also hope the remarks on the inaccessibility of procurement remedies for the contracting authorities that have infringed EU procurement law will feature in the judgment. All of this will provide helpful clarity on issues that should be uncontroversial under general EU law, but which seem to be susceptible of fueling litigation at domestic level.

Did you use AI to write this tender? What? Just asking! -- Also, how will you use AI to deliver this contract?

The UK’s Cabinet Office has published procurement policy note 2/24 on ‘Improving Transparency of AI use in Procurement’ (the ‘AI PPN’) because ‘AI systems, tools and products are part of a rapidly growing and evolving market, and as such, there may be increased risks associated with their adoption … [and therefore] it is essential to take steps to identify and manage associated risks and opportunities, as part of the Government’s commercial activities’.

The crucial risk the AI PPN seems to be concerned with relates to generative AI ‘hallucinations’, as it includes background information highlighting that:

‘Content created with the support of Large Language Models (LLMs) may include inaccurate or misleading statements; where statements, facts or references appear plausible, but are in fact false. LLMs are trained to predict a “statistically plausible” string of text, however statistical plausibility does not necessarily mean that the statements are factually accurate. As LLMs do not have a contextual understanding of the question they are being asked, or the answer they are proposing, they are unable to identify or correct any errors they make in their response. Care must be taken both in the use of LLMs, and in assessing returns that have used LLMs, in the form of additional due diligence.’

The PPN has the main advantage of trying to tackle the challenge of generative AI in procurement head on. It can help raise awareness in case someone was not yet talking about this and, more seriously, it includes an Annex A that brings together the several different bits of guidance issued by the UK government to date. However, the AI PPN does not elaborate on any of that guidance and is thus as limited as the Guidelines for AI procurement (see here), relatively complicated in that it points to rather different types of guidance ranging from ethics, to legal, to practical considerations, and requires significant knowledge and expertise to be operationalised (see here). Perhaps the best evidence of the complexity of the mushrooming sets of guidance is that the PPN itself includes in Annex A a reference to the January 2024 Guidance to civil servants on use of generative AI, which has been superseded by the Generative AI Framework for HMG, to which it also refers in Annex A. In other words, the AI PPN is not a ‘plug-and-play’ document setting out how to go about dealing with AI hallucinations and other risks in procurement. And given the pace of change in this area, it is also bound to be a PPN that requires multiple revisions and adaptations going forward.

A screenshot showing that the January guidance on generative AI use has been superseded (taken on 26 March 2024 10:20am).

More generally, the AI PPN is bound to be controversial and has already spurred insightful discussion on LinkedIn. I would recommend the posts by Kieran McGaughey and Ian Makgill. I offer some additional thoughts here and look forward to continuing the conversation.

In my view, one of the potential issues arising from the AI PPN is that it aims to cover quite a few different aspects of AI in procurement, as well as neglecting others. Slightly simplifying, there are three broad areas of AI-procurement interaction. First, there is the issue of buying AI-based solutions or services. Second, there is the issue of tenderers using (generative) AI to write or design their tenders. Third, there is the issue of the use of AI by contracting authorities, eg in relation to qualitative selection/exclusion, or evaluation/award decisions. The AI PPN covers aspects of . However, it is not clear to me that these can be treated together, as they pose significantly different policy issues. I will try to disentangle them here.

Buying and using AI

Although it mainly cross-refers to the Guidelines for AI procurement, the AI PPN includes some content relevant to the procurement and use of AI when it stresses that ‘Commercial teams should take note of existing guidance when purchasing AI services, however they should also be aware that AI and Machine Learning is becoming increasingly prevalent in the delivery of “non-AI” services. Where AI is likely to be used in the delivery of a service, commercial teams may wish to require suppliers to declare this, and provide further details. This will enable commercial teams to consider any additional due diligence or contractual amendments to manage the impact of AI as part of the service delivery.’ This is an adequate and potentially helpful warning. However, as discussed below, the PPN suggests a way to go about it that is in my view wrong and potentially very problematic.

AI-generated tenders

The AI PPN is however mostly concerned with the use of AI for tender generation. It recognises that there ‘are potential benefits to suppliers using AI to develop their bids, enabling them to bid for a greater number of public contracts. It is important to note that suppliers’ use of AI is not prohibited during the commercial process but steps should be taken to understand the risks associated with the use of AI tools in this context, as would be the case if a bid writer has been used by the bidder.’ It indicates some potential steps contracting authorities can take, such as:

• ‘Asking suppliers to disclose their use of AI in the creation of their tender.’

• ‘Undertaking appropriate and proportionate due diligence:

  • If suppliers use AI tools to create tender responses, additional due diligence may be required to ensure suppliers have the appropriate capacity and capability to fulfil the requirements of the contract. Such due diligence should be proportionate to any additional specific risk posed by the use of AI, and could include site visits, clarification questions or supplier presentations.

  • Additional due diligence should help to establish the accuracy, robustness and credibility of suppliers’ tenders through the use of clarifications or requesting additional supporting documentation in the same way contracting authorities would approach any uncertainty or ambiguity in tenders.’

• ‘Potentially allowing more time in the procurement to allow for due diligence and an increase in volumes of responses.’

• ‘Closer alignment with internal customers and delivery teams to bring greater expertise on the implications and benefits of AI, relative to the subject matter of the contract.’

In my view, there are a few problematic aspects here. While the AI PPN seems to try not to single out the use of generative AI as potentially problematic by equating it to the possible use of (human) bid writers, this is unconvincing. First, because there is (to my knowledge) no guidance whatsoever on an assessment of whether bid writers have been used, and because the AI PPN itself does not require disclosure of the engagement of bid writers (o puts any thought on the fact that third-party bid writers ma have used AI without this being known to the hiring tenderer, which would then require an extension of the disclosure of AI use further down the tender generation chain). Second, because the approach taken in the AI PP seems to point at potential problems with the use of (external, third-party) bid writers, whereas it does not seem to object to the use of (in-house) bid writers, potentially by much larger economic operators, which seems to presumptively not generate issues. Third, and most importantly, because it shows that perhaps not enough has been done so far to tackle the potential deceit or provision of misleading information in tenders if contracting authorities must now start thinking about how to get expert-based analysis of tenders, or develop fact-checking mechanisms to ensure bids are truthful. You would have thought that regardless of the origin of a tender, contracting authorities should be able to check their content to an adequate level of due diligence already.

In any case, the biggest issue with the AI PPN is how it suggests contracting authorities should deal with this issue, as discussed below.

AI-based assessments

The AI PPN also suggests that contracting authorities should be ‘Planning for a general increase in activity as suppliers may use AI to streamline or automate their processes and improve their bid writing capability and capacity leading to an increase in clarification questions and tender responses.’ One of the possibilities could be for contracting authorities to ‘fight fire with fire’ and also deploy generative AI (eg to make summaries, to scan for errors, etc). Interestingly, though, the AI PPN does not directly refer to the potential use of (generative) AI by contracting authorities.

While it includes a reference in Annex A to the Generative AI framework for HM Government, that document does not specifically address the use of generative AI to manage procurement processes (and what it says about buying generative AI is redundant given the other guidance in the Annex). In my view, the generative AI framework pushes strongly against the use of AI in procurement when it identifies a series of use cases to avoid (page 18) that include contexts where high-accuracy and high-explainability are required. If this is the government’s (justified) view, then the AI PPN has been a missed opportunity to say this more clearly and directly.

The broader issue of confidential, classified or proprietary information

Both in relation to the procurement and use of AI, and the use of AI for tender generation, the AI PPN stresses that it may be necessary:

• ‘Putting in place proportionate controls to ensure bidders do not use confidential contracting authority information, or information not already in the public domain as training data for AI systems e.g. using confidential Government tender documents to train AI or Large Language Models to create future tender responses.‘; and that

• ‘In certain procurements where there are national security concerns in relation to use of AI by suppliers, there may be additional considerations and risk mitigations that are required. In such instances, commercial teams should engage with their Information Assurance and Security colleagues, before launching the procurement, to ensure proportionate risk mitigations are implemented.’

These are issues that can easily exceed the technical capabilities of most contracting authorities. It is very hard to know what data has been used to train a model and economic operators using ‘off-the-shelf’ generative AI solutions will hardly be in a position to assess themselves, or provide any meaningful information, to contracting authorities. While there can be contractual constraints on the use of information and data generated under a given contract, it is much more challenging to assess whether information and data has been inappropriately used at a different link of increasingly complex digital supply chains. And, in any case, this is not only an issue for future contracts. Data and information generated under contracts already in place may not be subject to adequate data governance frameworks. It would seem that a more muscular approach to auditing data governance issues may be required, and that this should not be devolved to the procurement function.

How to deal with it? — or where the PPN goes wrong

The biggest weakness in the AI PPN is in how it suggests contracting authorities should deal with the issue of generative AI. In my view, it gets it wrong in two different ways. First, by asking for too much non-scored information where contracting authorities are unlikely to be able to act on it without breaching procurement and good administration principles. Second, by asking for too little non-scored information that contracting authorities are under a duty to score.

Too much information

The AI PPN includes two potential (alternative) disclosure questions in relation to the use of generative AI in tender writing (see below Q1 and Q2).

I think these questions miss the mark and expose contracting authorities to risks of challenge on grounds of a potential breach of the principle of equal treatment and the duty of good administration. The potential breach of the duty of good administration could be on grounds that the contracting authority is taking irrelevant information into account in the assessment of the relevant tender. The potential breach of equal treatment could come if tenders with some AI-generated elements were subjected to significantly more scrutiny than tenders where no AI was used. Contracting authorities should subject all tenders to the same level of due diligence and scrutiny because, at the bottom of it, there is no reason to ‘take a tenderer at its word’ when no AI is used. That is the entire logic of the exclusion, qualitative selection and evaluation processes.

Crucially, though, what the questions seem to really seek to ascertain is that the tenderer has checked for and confirms the accuracy of the content of the tender and thus makes the content its own and takes responsibility for it. This could be checked generally by asking all tenderers to confirm that the content of their tenders is correct and a true reflection of their capabilities and intended contractual delivery, reminding them that contracting authorities have tools to sanction economic operators that have ‘negligently provided misleading information that may have a material influence on decisions concerning exclusion, selection or award’ (reg.57(8)(i)(ii) PCR2015 and sch.7 13(2)(b) PA2023). And then enforcing them!

Checking the ‘authenticity’ of tenders when in fact contracting authorities are meant to check their truthfulness, accuracy and deliverability would be a false substitution of the relevant duties. It would also potentially eschew the incentives to disclose use of AI generation (lest contracting authorities find a reliable way of identifying it themselves and start applying the exclusion grounds above)—as thoroughly discussed in the LinkedIn posts referred to above.

too little information

Conversely, the PPN takes too soft and potentially confusing an approach to the use of AI to deliver the contract. The proposed disclosure question (Q3) is very problematic. It presents as ‘for information only’ a request for information on the use of AI or machine learning in the context of the actual delivery of the contract. This is information that will either relate to the technical specifications, award criteria or performance clauses (or all of them) and there is no meaningful way in which AI could be used to deliver the contract without this having an impact on the assessment and evaluation of the tender. The question is potentially misleading not only because of the indication that the information would not be scored, but also because it suggests that the use of AI in the delivery of a service or product is within the discretion of the tenderers. In my view, this would only be possible if the technical specifications were rather loosely written in performance terms, which would then require a very thorough description and assessment of how that performance is to be achieved. Moreover, the use of AI would probably require a set of organisational arrangements that should also not go unnoticed or unchecked in the procurement process. Moreover, one of the main challenges may not be in the use of AI in new contracts (were tenderers are likely to highlight it to stress the advantages, or to justify that their tenders are not abnormally low in comparison with delivery through ‘manual’ solutions), but in relation to pre-existing contracts. It also seems that a broader policy, recommendation and audit of the use of generative AI for the delivery of existing contracts and its treatment as a (permissible??) contract modification would have been needed.

Final thought

The AI PPN is an interesting development and will help crystallise many discussions that were somehow hovering in the background. However, a significant rethink is needed and, in my view, much more detailed guidance is needed in relation to the different dimensions of the interaction between AI and procurement. There are important questions that remain unaddressed and, in my view, one of the most pressing ones concerns the balance between general regulation and the use of procurement to regulate AI use. While the UK government remains committed to its ‘pro-innovation’ approach and no general regulation of AI use is put in place, in particular in relation to public sector AI use, procurement will continue to struggle and fail to act as a regulator of the technology.

The principle of competition is dead. Long live the principle of competition (recording)

Here is the recording for the first part of today’s seminar ‘The principle of competition is dead. Long live the principle of competition’, with renewed thanks to Roberto Caranta, Trygve Harlem Losnedahl and Dagne Sabockis for sharing their great insights. A transcript is available here, as well as the slides I used. As always, comments welcome and more than happy to continue the discussion!

Transposing Directives no longer so discretionary! The Court of Justice forces transposition of discretionary exclusion grounds and hints at ‘intra-State’ vertical direct effect (C‑66/22)

** This comment was first published as an Op-Ed for EU Law Live on 8 December 2022 (see formatted version). I am reposting it here in case of broader interest. **

On the face of it, in Infraestruturas de Portugal and Futrifer Indústrias Ferroviárias (C-66/22), the Court of Justice had to assess whether Member States can limit the exclusion of competition law violators from participation in tenders for public contracts to cases where the national competition authority has previously imposed such debarment as an ancillary penalty. While this is a plausible transposition approach that seeks to centralise competition law analysis under the control of the specialist administrative authority, it can also reduce the effectiveness of procurement mechanisms seeking to preserve (much needed) competition for public contracts. It is thus fair enough to test the boundaries of the discretion that contracting authorities can retain in this context. However, in Infraestruturas, the Court of Justice did two other things that are potentially significant beyond the narrower field of procurement governance. First, the Court reversed its previous case law and established that Member States are (no longer) allowed not to transpose discretionary exclusion grounds. This says something (but I am not too sure what) about the more general level of discretion that Member States retain in the transposition of (prescriptively worded) Directives into their national systems under Art 288 TFEU. Second, the Court furthered a line of reasoning that comes to assign ‘individual rights’ to contracting authorities—that is, entities within the public sector—in what could seem like the creation of an ‘intra-State’ modality of vertical direct effect. In this Op-Ed, I try to make some sense of these two developments and leave aside for now the details of the interpretation of the specific grounds for the exclusion of economic operators under Directive 2014/24/EU.

No longer discretionary to transpose discretionary exclusion grounds

It is trite EU law that, under Art 288 TFEU, Member States retain discretion in the choice of form and methods for the transposition of a Directive. While Directives can be prescriptive about their aims and goals and, sometimes, about specific modes of protection of the relevant legal interest, there seems to be a (theoretical) agreement that Directives still (must) leave a margin of discretion to Member States—else, they surreptitiously shapeshift into Regulations. Such discretion would seem to cover in particular those elements of a Directive that are explicitly labelled as discretionary. This ‘orthodoxy’ seemed to be straightforwardly applied by the Court of Justice in the analysis of the constraints on the transposition of the Public Procurement Directive 2014/24.

The Public Procurement Directive contains a set of rules on the exclusion from tenders for public contracts of economic operators that have fallen short of their legal obligations. In Art. 57, in addition to setting rules applicable to all exclusion decisions, the Directive distinguishes between, on the one hand, mandatory exclusion grounds that require contracting authorities to exclude economic operators convicted by final judgment for one of a series of breaches (Art. 57(1)) and, on the other hand, discretionary exclusion grounds that allow contracting authorities to exclude the affected economic operators (Art. 57(4)). Member States are explicitly allowed to turn discretionary exclusion grounds mandatory under their transposing legislation. Conversely, until now, the Court had been clear that Member States were allowed not to transpose discretionary exclusion grounds. So far, so good.

In Infraestruturas, however, the Court of Justice U-turned. It stated that:

… the first subparagraph of Article 57(4) of Directive 2014/24 … states that ‘contracting authorities may exclude or may be required by Member States to exclude any economic operator from participation in a procurement procedure’ in any of the situations referred to in points (a) to (i) of that provision.

In that connection, it admittedly follows from certain judgments of the Court … that the Member States can decide whether or not to transpose the facultative grounds for exclusion referred to in that provision. The Court has in fact held that … the Member States are free not to apply the facultative grounds for exclusion set out in that directive or to incorporate them into national law with varying degrees of rigour according to legal, economic or social considerations prevailing at national level (see, to that effect, judgments of 19 June 2019, Meca, C‑41/18, EU:C:2019:507, paragraph 33; of 30 January 2020, Tim, C‑395/18, EU:C:2020:58, paragraphs 34 and 40; and of 3 June 2021, Rad Service and Others, C‑210/20, EU:C:2021:445, paragraph 28).

However, an analysis of the wording of the first subparagraph of Article 57(4) of Directive 2014/24, the context into which that provision fits, and the aim that the latter pursues within the framework of that directive, shows that contrary to what is apparent from those judgments, the Member States are under the obligation to transpose that provision into their national law (C-66/22, paras. 48-50, emphases added).

In my view, this U-turn challenges the ‘orthodoxy’ to the extent that the Court subjects the margin of discretion left to the Member States by the EU legislators to the Court’s assessment of whether what is clearly labelled as discretionary—and was as such treated in earlier case law—is permissibly left to the discretion of the Member States in view of the aims of the Directive. I think that this introduces a potentially tricky line of challenge of the content of EU Directives on the grounds that the EU legislators could not have left to the Member States’ discretion specific aspects of their content without undermining the goals of the Directive itself. This can ultimately constrain the upstream discretion in the choice of legal instrument under Art 288 TFEU by the EU legislators themselves, and further erode the distinction between Regulations and Directives if the content of the Directives can in fact eventually be binding in their entirety and directly applicable in all Member States. Further, this U-turn is based on a rather peculiar interpretation of the wording of the Public Procurement Directive that comes to assign ‘individual rights’ to the public sector. Given this peculiarity, I am not too sure whether the deviation from the orthodoxy in Infraestruturas indicates a significant shift by the Court of Justice or ‘just’ an exception or oddity that may confirm the general rule.

‘Intra-State’ vertical direct effect?

In justifying its U-turn, the Court of Justice stresses that, under Art. 57(4) of the Public Procurement Directive:

… the choice as to the decision whether or not to exclude an economic operator from a public procurement procedure on one of the grounds set out in that provision falls to the contracting authority, unless the Member States decide to transform that option to exclude into an obligation to do so. Accordingly, the Member States must transpose that provision either by allowing or by requiring contracting authorities to apply the exclusion grounds laid down by the latter provision. … a Member State cannot omit those grounds from its national legislation transposing Directive 2014/24 and thus deprive contracting authorities of the possibility – which must, at the very least, be conferred on them by virtue of that provision – of applying those grounds.

… it should be noted that recital 101 of that directive states that ‘contracting authorities should … be given the possibility to exclude economic operators which have proven unreliable’. That recital thus confirms that a Member State must transpose that provision in order not to deprive contracting authorities of the possibility referred to in the preceding paragraph and that recital.

…

Lastly, as to the objective pursued by Directive 2014/24 in so far as concerns the facultative grounds for exclusion, the Court has acknowledged that that objective is reflected in the emphasis placed on the powers of contracting authorities. Thus the EU legislature intended to confer on the contracting authority, and on it alone, the task of assessing whether a candidate or tenderer must be excluded from a procurement procedure during the stage of selecting the tenderers (see, to that effect, judgments of 19 June 2019, Meca, C‑41/18, EU:C:2019:507, paragraph 34, and of 3 October 2019, Delta Antrepriză de Construcţii şi Montaj 93, C‑267/18, EU:C:2019:826, paragraph 25).

The option, or indeed obligation, for the contracting authority to apply the exclusion grounds set out in the first subparagraph of Article 57(4) of Directive 2014/24 is specifically intended to enable it to assess the integrity and reliability of each of the economic operators participating in a public procurement procedure.

The EU legislature thus intended to ensure that contracting authorities have, in all Member States, the possibility of excluding economic operators who are regarded as unreliable by those authorities (C-66/22, paras. 51-52 and 55-57, emphases added).

Even if not altogether new—see Meca (C-41/18) and Delta (C-267/18)—I find this line of reasoning puzzling. The way the Court of Justice has interpreted Art. 57(4) of the Public Procurement Directive equates to an ‘individual right’ for contracting authorities not to contract with economic operators they deem unreliable and, crucially, this is an ‘individual right’ that Member States cannot deprive them from. The protection of such right implies an ‘intra-State’ modality of vertical direct effect—at least to the extent that, after Infraestruturas, a contracting authority of any Member State with centralised exclusion decision-making can challenge any constraints on its administrative discretion and simply set aside the domestic rules and directly rely on the Directive to proceed to exclusion on the basis of discretionary grounds.

To my mind, this line of reasoning extracts the wrong implications from the wording of the Directive because of the quasi-anthropomorphism of contracting authorities. Given that the Directive conceptualises contracting authorities as the relevant unit of decision-making, references to contracting authorities should be seen as references to decisions within a procurement procedure, not as references to agents that derive rights independently from—or even against—the structure of the State into which they are embedded. In the end, contracting authorities are defined as ‘the State, regional or local authorities, bodies governed by public law or associations formed by one or more such authorities or one or more such bodies governed by public law’ (Art. 2(1)(1) Directive 2014/24). A functional interpretation of the wording of Article 57(4) of the Public Procurement Directive would recognise that the meaning of ‘contracting authorities may exclude or may be required by Member States to exclude’ is that ‘in a covered procurement procedure, it is permissible to exclude, and it can be made mandatory to exclude’—which would then straightforwardly follow the orthodoxy in allowing Member States to exercise the discretion on the form and method of transposition of that possibility.

I submit that the Court of Justice has followed a line of reasoning that is also problematic in relation to other provisions of the Public Procurement Directive, in particular in relation to the potential effects it could have in ‘empowering’ contracting authorities to take courses of action (eg international collaboration) that could imply domestic ultra vires.

Final thoughts

What I find most confusing in this part of the Infraestruturas Judgment is that the Court could have found much less disruptive and confusing ways to reach the same conclusion. For example, it could have found that, in empowering the national competition authority to make decisions on the exclusion of tenderers through the imposition of ancillary penalties, Portugal had decided to transpose the relevant discretionary exclusion ground, but done so incorrectly or defectively by simultaneously transposing the ground but limiting the discretion of the contracting authority. I would still find issue with that approach, but at least it would be easier to reconcile with important parts of the orthodoxy of fundamental aspects of EU law.

Will the ECJ mandate protectionism in procurement -- comments on AG Collins' Kolin Opinion (C-652/22)

In the Opinion in Kolin Inşaat Turizm Sanayi ve Ticaret (C-652/22, EU:C:2024:212, hereafter ‘Kolin’), Advocate General Collins has argued that only economic operators established in countries party to international agreements on public contracts that bind the EU may rely on the provisions of Directive 2014/25/EU. This would imply that economic operators established in other countries are not entitled to participate in a public contract award procedure governed by Directive 2014/25/EU and, consequently, are unable to rely on the provisions of that Directive before Member State courts. In my view, this interpretation is both protectionist and problematic. The ECJ should not follow it. In this blog I sketch the reasons for this view.

Limited (international law) obligation of equal treatment does not imply a general (EU) obligation to exclude or discriminate

The Kolin Opinion concerns the interpretation of Art 43 of Directive 2014/25/EU, which could be relevant to the interpretation of Art 25 of Directive 2014/24/EU (on which see A La Chimia, ‘Article 25’ in R Caranta and A Sanchez-Graells (eds), European Public Procurement. Commentary on Directive 2014/24/EU (Edward Elgar 2021) 274-286. Art 43 of Dir 2014/25/EU establishes that:

In so far as they are covered by Annexes 3, 4 and 5 and the General Notes to the European Union’s Appendix I to the GPA and by the other international agreements by which the Union is bound, contracting entities within the meaning of Article 4(1)(a) shall accord to the works, supplies, services and economic operators of the signatories to those agreements treatment no less favourable than the treatment accorded to the works, supplies, services and economic operators of the Union.

AG Collins considers that

It follows that economic operators from non-covered third-countries do not fall within the scope ratione personae of Directive 2014/25 ... Since the applicant is not entitled to participate in a procedure for the award of a public contract governed by Directive 2014/25, it cannot seek to rely on the provisions thereof before a Member State court. The referring court therefore cannot obtain a response to a reference for a preliminary ruling on the interpretation of those provisions, since any answer that the Court might give to its request would not have binding effect. That reason suffices to justify a finding that this reference for a preliminary ruling is inadmissible (para 33, emphases added).

This position implies a logical jump in the reasoning. While it is plain that only economic operators from covered third-countries have a legally enforceable right to participate in public tenders on equal terms, it is by no means clear that other economic operators must necessarily be excluded from participation. If that was the plain interpretation and implication of that provision (and Art 25 Dir 2014/24/EU), the Commission would not have needed to develop the International Procurement Instrument (IPI) to establish circumstances in which exclusion of economic operators from non-covered third countries can be mandated. Along the same lines, AG Rantos argued in the Opinion in CRRC Qingdao Sifang and Others (C-266/22, EU:C:2023:399, not available in English) that ‘Member States can grant less favourable treatment to economic operators from non-covered third parties’ (para 65, own translation from Spanish).

In fact, as the Opinion reflects, ‘[a]lmost all of the parties to the procedure before the Court take the view that Member States may regulate the participation of economic operators from third-countries in procedures for the award of public contracts’ (para 35). In particular, the Croatian government submitted ‘that EU law contains no general prohibition on the participation of economic operators from third-countries in procedures for the award of public contracts in the European Union’ and provided sound arguments in support of that, as the ‘Commission’s Guidance on the participation of third-country bidders confirms that proposition where it states that economic operators from third-countries may be excluded from these procedures, without requiring their exclusion’ (para 36). Those arguments are also aligned with AG Rantos’ CRRC Opinion (paras 72-74). Estonia also submitted there is no obligation under EU law to limit participation by economic operators from non-covered third parties (para 38). Denmark, France, and Austria also considered that there is no ban stemming from EU law, even if the Union has exclusive competence in relation to the common commercial policy (paras 39-40). This should have given the AG pause.

Instead, as suggested by the Commission, AG Collins seeks to support the Opinion’s logical jump in an additional legal argument based on the remit of the EU’s competence to regulate the participation of economic operators from third-countries in procurement procedures in the European Union. The key issue here is not whether the EU has an exclusive or a shared competence in procurement, but that AG Collins considers that

… by adopting Article 43 of Directive 2014/25, the European Union has exercised its competence in relation to economic operators established in a country party to the GPA or to another international agreement on the award of public contracts by which the European Union is bound. … economic operators established in Türkiye do not come within that category. Although the European Union has not exercised its exclusive competence to establish whether economic operators from non-covered third-countries may participate in such procedures, Member States may not rely on that fact in order to regain competence to act in that area (para 50, emphases added).

Since the European Union does not appear to have exercised its exclusive competence to determine access by economic operators from non-covered third-countries to procedures for the award of public contracts, Member States wishing to take steps to that end may inform the competent EU institutions of their proposed course of action with a view to obtaining the requisite authorisation. Nothing in the Court’s file indicates that Croatia has taken such a step. Second, unilateral action by Member States could undermine the European Union’s bargaining position in the context of its efforts to open, on a reciprocal basis, markets for public contracts in third countries. Third, it could interfere with the uniform application of EU law, since in such circumstances the application of Directive 2014/25 ratione personae could vary from one Member State to another. (para 52, emphases added).

In my view, AG Collins is conflating normative and positive analysis. It is clear that dissimilar approaches in the Member States undermine the Commission’s bargaining position—thus the need to bring in the IPI as well as other instruments such as the Foreign Subsidies Regulation (FSR)—and can lead an absence of uniformity in the application of the Directive. However, these are normative issues. From a positive standpoint, it is in my view incorrect to state that the EU has exercised its competence in relation to GPA and other covered third country operators through Article 43 of Directive 2014/25/EU or, for that matter, Article 25 of Directive 2014/24/EU. The exercise of the relevant competence concerns the entering into those international treaties, not the internal legislative measures put in place to promote their effectiveness.

To me, it is clear that the obligation to grant equal treatment to GPA and other covered economic operators stems from the GPA or the relevant international treaty, not the Directive. Art 43 Dir 2014/25/EU (and Art 25 Dir 2014/24/EU are mere reminders of the obligations under international law and cannot alter them. Their omission would not have made any difference in covered third country economic operators’ legal position. By the same token, their inclusion cannot serve to prejudice the position of non-covered third country economic operators. As above, the whole process leading to the IPI and FSR would have been entirely superfluous. In my view, the Kolin Opinion follows too closely the dangerously protectionist policy approach pushed by the Commission, and does so in a way that is not legally persuasive (or accurate, in my view).

‘Tolerance’ of third country economic operators’ participation must engage legal protection under the CFR

Moreover, the Kolin Opinion would open a very dangerous path in terms of rule of law and upholding the effectiveness of the Charter of Fundamental Rights, especially Articles 41 and 47—and allow contracting authorities two bites of the cherry in relation to tenders submitted by economic operators from non-covered third countries. Contracting authorities could ‘tolerate’ participation of non-covered third country economic operators to see if those are the ones providing the most economically advantageous offer and, if not, or if other (industrial policy) considerations kicked in, they could simply reject or set aside the tender. This would happen in a context of insufficient guarantees.

Even assuming there was an obligation to exclude under Art 43 of Directive 2014/28/EU or Art 25 of Directive 2014/24/EU, which there is not, contracting authorities would be bound by the duties under Art 41 CFR in relation to EU and covered third-country economic operators. The relevant duty would require an immediate exclusion of the not covered economic operators to protect the (in that case) participation rights of those covered. A contracting authority that had not carried out such exclusion could seek to benefit from the advantages provided by the third country economic operator in breach of its duties, which is not permissible.

Conversely, a contracting authority that had not discharged its duty to exclude would be allowed to still benefit from its inaction by discriminating against and eventually excluding at a later stage the tender of the third-country economic operator without the latter having legal recourse. This would also not be in line with the effectiveness of Arts 41 and 47 CFR and certainly not in line with the doctrine of legitimate expectations. Once an economic operator or tenderer is not excluded or rejected at the first opportunity, there is a positive and very specific representation made by the contracting authority that the economic operator and/or its tender is in the run for the contract. This must trigger legal protection—although the specific form is likely to depend on domestic administrative law.

In the case at hand, like in many other cases in daily practice, despite Kolin not being eligible for equal treatment under Art 43 of Directive 2014/24/EU—and thus not having an enforceable right to participate in the tender and to equal treatment within it deriving from international law—the contracting authority had ‘tolerated’ its participation. The Opinion is plain that, following the receipt of tenders, the contracting authority ‘concluded that 6 out of the 15 tenders submitted fulfilled the selection criteria. [Kolin], a company established in Türkiye, submitted one of the tenders selected’ (para 16). However, the Opinion does not grant Kolin any rights because of such tolerance.

Contrary to the view held by AG Collins, ‘The Austrian Government contends that, although, in principle, Directive 2014/25 does not apply to economic operators from non-covered third-countries, such operators may rely on that directive once a contracting authority has permitted their participation in a procedure for the award of a public contract award’ (para 26). I share this view. Crucially, this is not an issue the Opinion explicitly addresses. But this is the main reason why the ECJ should not follow the Opinion.

Initial UK guidance on pro-innovation AI regulation: Much ado about nothing?

The UK Government’s Department for Science, Innovation and Technology (DSIT) has recently published its Initial Guidance for Regulators on Implementing the UK’s AI Regulatory Principles (Feb 2024) (the ‘AI guidance’). This follows from the Government’s response to the public consultation on its ‘pro-innovation approach’ to AI regulation (see here).

The AI guidance is meant to support regulators develop tailored guidance for the implementation of the five principles underpinning the pro-innovation approach to AI regulation, that is: (i) Safety, security & robustness; (ii) Appropriate transparency and explainability; (iii) Fairness;
(iv) Accountability and governance; and (v) Contestability and redress.

Voluntary approach and timeline for implementation

A first, perhaps, surprising element of the AI guidance comes from the way in which engagement with the principles by current regulators is framed as voluntary. The white paper describing the pro-innovation approach to AI regulation (the ‘AI white paper’) had indicated that, initially, ‘the principles will be issued on a non-statutory basis and implemented by existing regulators’, with a clear expectation for regulators to make use their ‘domain-specific expertise to tailor the implementation of the principles to the specific context in which AI is used’.

The AI white paper made it clear that a failure by regulators to implement the principles would lead the government to introduce ‘a statutory duty on regulators requiring them to have due regard to the principles’, which would still ‘allow regulators the flexibility to exercise judgement when applying the principles in particular contexts, while also strengthening their mandate to implement them’. There seemed to be little room for discretion for regulators to decide whether to engage with the principles, even if they were expected to exercise discretion on how to implement them.

By contrast, the initial AI guidance indicates that it ‘is not intended to be a prescriptive guide on implementation as the principles are voluntary and how they are considered is ultimately at regulators’ discretion’. There is also a clear indication in the response to the public consultation that the introduction of a statutory duty is not in the immediate legislative horizon and the absence of a pre-determined date for the assessment of whether the principles have been ‘sufficiently implemented’ on a voluntary basis (for example, in two years’ time) will make it very hard to press for such legislative proposal (depending on the policy direction of the Government at the time).

This seems to follow from the Government’s position that ‘acknowledge[s] concerns from respondents that rushing the implementation of a duty to regard could cause disruption to responsible AI innovation. We will not rush to legislate’. At the same time, however, the response to the public consultation indicates that DSIT has asked a number of regulators to publish by 30 April 2024 updates on their strategic approaches to AI. This seems to create an expectation that regulators will in fact engage—or have defined plans for engaging—with the principles in the very short term. How this does not create a ‘rush to implement’ and how putting the duty to consider the principles on a statutory footing would alter any of this is hard to fathom, though.

An iterative, phased approach

The very tentative approach to the issuing of guidance is also clear in the fact that the Government is taking an iterative, phased approach to the production of AI regulation guidance, with three phases foreseen. A phase one consisting of the publication of the AI guidance in Feb 2024, a phase two comprising an iteration and development of the guidance in summer of 2024, and a phase three (with no timeline) involving further developments in cooperation with regulators—to eg ‘encourage multi-regulator guidance’. Given the short time between phases one and two, some questions arise as to how much practical experience will be accumulated in the coming 4-6 months and whether there is much value in the high-level guidance provided in phase one, as it only goes slightly beyond the tentative steer included in the AI white paper—which already contained some indication of ‘factors that government believes regulators may wish to consider when providing guidance/implementing each principle’ (Annex A).

Indeed, the AI guidance is still rather high-level and it does not provide much substantive interpretation of what the different principles mean. It is very much a ‘how to develop guidance’ document, rather than a document setting out core considerations and requirements for regulators to embed within their respective remits. A significant part of the document provides guidance on ‘interpreting and applying the AI regulatory framework’ (pp 7-12) but this is really ‘meta-guidance’ on issues such as potential collaboration between regulators for the issuance of joint guidance/tools, or an encouragement to benchmarking and the avoidance of duplicated guidance where relevant. General recommendations such as the value of publishing the guidance and keeping it updated seem superfluous in a context where the regulatory approach is premised on ‘the expertise of [UK] world class regulators’.

The core of the AI guidance is limited to the section on ‘applying individual principles’ (pp 13-22), which sets out a series of questions to consider in relation to each of the five principles. The guidance offers no answers and very limited steer for their formulation, which is entirely left to regulators. We will probably have to wait (at least) for the summer iteration to get some more detail of what substantive requirements relate to each of the principles. However, the AI guidance already contains some issues worthy of careful consideration, in particular in relation to the tunnelling of regulatory power and the imbalanced approach to the different principles that follows from its reliance on existing (and soon to emerge) technical standards.

technical standards and interpretation of the regulatory principles

regulatory tunnelling

As we said in our response to the public consultation on the AI white paper,

The principles-based approach to AI regulation suggested in the AI [white paper] is undeliverable, not only due to lack of detail on the meaning and regulatory implications of each of the principles, but also due to barriers to translation into enforceable requirements, and tensions with existing regulatory frameworks. The AI [white paper] indicates in Annex A that each regulator should consider issuing guidance on the interpretation of the principles within its regulatory remit, and suggests that in doing so they may want to rely on emerging technical standards (such as ISO or IEEE standards). This presumes both the adequacy of those standards and their sufficiency to translate general principles into operationalizable and enforceable requirements. This is by no means straightforward, and it is hard to see how regulators with significantly limited capabilities … can undertake that task effectively. There is a clear risk that regulators may simply rely on emerging industry-led standards. However, it has already been pointed out that this creates a privatisation of AI regulation and generates significant implicit risks (at para 27).

The AI guidance, in sticking to the same approach, confirms this risk of regulatory tunnelling. The guidance encourages regulators to explicitly and directly refer to technical standards ‘to support AI developers and AI deployers’—while at the same time stressing that ‘this guidance is not an endorsement of any specific standard. It is for regulators to consider standards and their suitability in a given situation (and/or encourage those they regulate to do so likewise).’ This does not seem to be the best approach. Leaving it to each of the regulators to assess the suitability of existing (and emerging) standards creates duplication of effort, as well as a risk of conflicting views and guidance. It would seem that it is precisely the role of centralised AI guidance to carry out that assessment and filter out technical standards that are aligned with the overarching regulatory principles for implementation by sectoral regulators. In failing to do that and pushing the responsibility down to each regulator, the AI guidance comes to abdicate responsibility for the provision of meaningful policy implementation guidelines.

Additionally, the strong steer to rely on references to technical standards creates an almost default position for regulators to follow—especially those with less capability to scrutinise the implications of those standards and to formulate complementary or alternative approaches in their guidance. It can be expected that regulators will tend to refer to those technical standards in their guidance and to take them as the baseline or starting point. This effectively transfers regulatory power to the standard setting organisations and further dilutes the regulatory approach followed in the UK, which in fact will be limited to industry self-regulation despite the appearance of regulatory intervention and oversight.

unbalanced approach

The second implication of this approach is that some principles are likely to be more developed than other in regulatory guidance, as they also are in the initial AI guidance. The series of questions and considerations are more developed in relation to principles for which there are technical standards—ie ‘safety, security & robustness’, and ‘accountability and governance’—and to some aspects of other principles for which there are standards. For example, in relation to ‘adequate transparency and explainability’, there is more of an emphasis on explainability than on transparency and there is no indication of how to gauge ‘adequacy’ in relation to either of them. Given that transparency, in the sense of publication of details on AI use, raises a few difficult questions on the interaction with freedom of information legislation and the protection of trade secrets, the passing reference to the algorithmic transparency recording standard will not be sufficient to support regulators in developing nuanced and pragmatic approaches.

Similarly, in relation to ‘fairness’, the AI guidance solely provides some reference in relation to AI ethics and bias, and in both cases in relation to existing standards. The document falls awfully short of any meaningful consideration of the implications and requirements of the (arguably) most important principle in AI regulation. The AI guidance solely indicates that

Tools and guidance could also consider relevant law, regulation, technical standards and assurance techniques. These should be applied and interpreted similarly by different regulators where possible. For example, regulators need to consider their responsibilities under the 2010 Equality Act and the 1998 Human Rights Act. Regulators may also need to understand how AI might exacerbate vulnerabilities or create new ones and provide tools and guidance accordingly.

This is unhelpful in many ways. First, ensuring that AI development and deployment complies with existing law and regulation should not be presented as a possibility, but as an absolute minimum requirement. Second, the duties of the regulators under the EA 2010 and HRA 1998 are likely to play a very small role here. What is crucial is to ensure that the development and use of the AI is compliant with them, especially where the use is by public sector entities (for which there is no general regulator—and in relation to which a passing reference to the EHRC guidance on AI use in the public sector will not be sufficient to support regulators in developing nuanced and pragmatic approaches). In failing to explicitly acknowledge the existence of approaches to the assessment of AI and algorithmic impacts on fundamental and human rights, the guidance creates obfuscation by omission.

‘Contestability and redress’ is the most underdeveloped principle in the AI guidance, perhaps because no technical standard addresses this issue.

final thoughts

In my view, the AI guidance does little to support regulators, especially those with less capability and resources, in their (voluntary? short-term?) task of issuing guidance in their respective remits. Meaningful AI guidance needs to provide much clearer explanations of what is expected and required for the correct implementation of the five regulatory principles. It needs to address in a centralised and unified manner the assessment of existing and emerging technical standards against the regulatory benchmark. It also needs to synthesise the multiple guidance documents issued (and to be issued) by regulators—which it currently simply lists in Annex 1—to avoid a multiplication of the effort required to assess their (in)comptability and duplications. By leaving all these tasks to the regulators, the AI guidance (and the centralised function from which it originates) does little to nothing to move the regulatory needle beyond industry-led self-regulation and fails to discharge regulators from the burden of issuing AI guidance.

High hopes but little movement for public sector AI use regulation through procurement in the UK Government's 'Pro-innovation Approach' response

The UK Government has recently published its official response (the ‘response’) to the public consultation of March 2023 on its ‘pro-innovation approach’ to AI regulation (for an initial discussion, see here). The response shows very little movement from the original approach and proposals and, despite claiming that significant developments have already taken place, it mainly provides a governmental self-congratulatory narrative and limited high-level details of a regulatory architecture still very much ‘under construction’. The publication of the response was coupled with that of Initial Guidance for Regulators on Implementing the UK’s AI Regulatory Principles (Feb 2024), which I will comment in a subsequent post.

A section of particular interest in the response refers to ‘Ensuring AI best practice in the public sector’ (at 21-22), which makes direct reference to the use of public procurement and the exercise of public sector buying power as a regulatory lever.

This section describes some measures being put in place or planned to seize ‘the opportunities presented by AI to deliver better public services including health, education, and transport’, such as:

• ‘tripling the number of technical AI engineers and developers within the Cabinet Office to create a new AI Incubator for the government’ (para 41).
  This is an interesting commitment to building in-house capability. It would however be interesting to know whether these are new or reassigned roles, as well as how the process of recruitment and retention is faring, given the massive difficulties evidenced in the recent analysis by the National Audit Office, Digital transformation in government: addressing the barriers to efficiency (10 Mar 2023, HC 2022-23, 1171).

• ‘The government is also using the procurement power of the public sector to drive responsible and safe AI innovation. The Central Digital and Data Office (CDDO) has published guidance on the procurement and use of generative AI for the UK government. Later this year, DSIT will launch the AI Management Essentials scheme, setting a minimum good practice standard for companies selling AI products and services. We will consult on introducing this as a mandatory requirement for public sector procurement, using purchasing power to drive responsible innovation in the broader economy’ (para 43).
  This is also an interesting aspiration, for several reasons. First, the GenAI guidance is very generic and solely highlights pre-existing (also very generic) guidance on how to carry out procurement of AI (see screenshot below). This can hardly be seen as a meaningful development of the existing regulatory framework. Second, the announcement of an ‘AI Management Essentials’ scheme seems to be mirrored on the ‘Cyber Essentials’ scheme in the area of cyber security, despite significant differences and the much higher level of complexity that can be expected from an ‘all-encompassing’ scheme for the management of the myriad risks generated by the deployment of AI.

Screenshot of the webpage https://www.gov.uk/government/publications/generative-ai-framework-for-hmg/generative-ai-framework-for-hmg-html (accessed 22 February 2024), where this information is available in accessible format.

• ‘This builds on the Algorithmic Transparency Recording Standard (ATRS), which established a standardised way for public sector organisations to proactively publish information about how and why they are using algorithmic methods in decision-making. Following a successful pilot of the standard, and publication of an approved cross-government version last year, we will now be making use of the ATRS a requirement for all government departments and plan to expand this across the broader public sector over time’ (para 44).
  This is also interesting in that the ‘success’ attributed to the development of the ATRS is very clearly undermined by the almost absolute lack of use other than in relation to the pilot projects (see screenshot below). It is also interesting that the ATRS allows public sector AI deployers to fill in but not publish the relevant documents, as a form of self-reflective/evaluative exercise. I wonder how many publications we will see in the coming months, even if ‘use of the ATRS’ becomes a requirement.

Screenshot of the list of published transparency disclosures at Algorithmic Transparency Reports - GOV.UK (www.gov.uk) (accessed 22 February 2024), where this information is available in accessible format.

Overall, I think the response to the ‘pro-innovation’ AI regulation consultation does little to back up the high expectations being placed in public procurement as a mechanism of regulation by contract. I will update the analysis in this UK-focused paper on the use of procurement to regulate public sector AI use before final publication, but there will be little change. The broader analysis in my recent monograph also remains applicable (phew): Digital Technologies and Public Procurement. Gatekeeping and experimentation in digital public governance (OUP 2024).