Competition Law Insights – Artificial Intelligence

Competition authorities across the globe have, like the vast majority of organisations, been investing significantly in building their understanding of AI and the potential competition and consumer law impacts of its development and use.

In this briefing, we discuss the ways in which competition authorities are seeking to tackle the key issues arising from AI through a competition law lens.

You can also listen to our podcast: AI Insights - Episode 6: The key issues related to AI through a competition law lens

Now Reading

CMA Reliance on Existing Toolkits

Given the economic principles underpinning competition law are adaptable across many industries and sectors, competition agencies (including the UK Competition & Markets Authority ('CMA'), the European Commission and the US Department of Justice among others) are focussed on the ways in which existing competition laws can be applied to AI markets and the challenges of AI. The UK has not, as yet, created new laws to capture or regulate AI (see our earlier briefing here).

Whilst much hype has been generated around the new Digital Markets, Competition and Consumers Act in the UK ('DMCC Act') and the Digital Markets Act in the EU, those pieces of legislation do not deal explicitly with AI. The UK CMA has, however, suggested that it is ready to use its new powers under the DMCC Act to "raise standards" in AI markets and to "tackle firms that do not play by the rules through enforcement action".

However, the EU’s AI Act will enter into force on 1 August 2024 (see our briefing here) which, whilst regulatory-focused, will touch on EU competition law enforcement. By way of example, the AI Act includes broad procedural powers for national supervisory agencies to access company data to check for compliance with the AI Act. Those national supervisory authorities are required by the Act to provide this data to competition authorities annually where that information "may be of potential interest" for the application of EU competition rules: a relatively low bar.

So, how has the UK CMA been looking at the AI space to date? The CMA has:

Called out AI as a key focus area in its Annual Report for 2024/2025.
Published various materials in relation to AI, including:
- A strategic update on the CMA’s approach to AI generally.
- An initial report and an update paper in connection with the CMA's initial review into foundation models.
- The CMA's first merger decision in relation to AI markets, in connection with the partnership between Microsoft and Mistral AI.
- The signing of a joint statement with the European Commission, US Federal Trade Commission and US Department of Justice setting out shared principles to support competition and protect consumers in the AI space.

The direction of travel is similar in the consumer law space, with the CMA's strategic update referring to plans to enforce existing legislation to protect consumers against unfair commercial practices and unfair terms.

The CMA's Initial Review Into Foundation Models

As part of its initial review into foundation models ('FMs'), the CMA identified a number of issues with the development of AI markets, the vast majority of which it could plausibly address using its existing antitrust enforcement powers under the Competition Act 1998, its merger control review powers and its market investigations powers. For example, in the abuse of dominance space, the CMA expressed concern that large incumbent technology firms (with existing power in other important digital markets) could be positioned to leverage that power to, for example, distort choice in FM services and restrict competition in FM deployment.

Further, with the aim of proactively seeking to stop anticompetitive conduct from arising, and to guide the sector toward positive competitive outcomes across the FMs value chain, the CMA also worked up some new "AI principles", based on conventional competition law principles. These principles are contained in the CMA's initial review into FMs and its strategic update.

These principles include (i) ensuring ongoing and ready access to inputs needed for the development of AI, such as data; (ii) maintaining fair dealing; and (iii) safeguarding transparency, for example, by ensuring that developers provide the right information to deployers to allow them to make informed choices.

Merger Control Scrutiny of AI Partnerships

Another notable point arising from the CMA's initial review into FMs and strategic update is the concern that partnerships involving key players in the AI value chain could reinforce or extend existing positions of market power. Existing merger control tools may well be of use to regulators in this respect, and we have seen the beginning of the CMA's interest in its Microsoft/Mistral decision.

Indeed, a number of competition authorities are taking a keen interest in partnerships between AI pure players and big tech. These partnerships are a prevalent feature of AI development across the globe, and have been highlighted as a potentially significant issue by both regulators and industry participants.

The question of whether any one of these partnerships may have a substantive impact on competition is yet to be looked at by any authority, but the debate around them, in very high-level terms, centres around whether:

(i) AI partnerships are good for competition, in that they enable pure players to access the inputs they need (such as investment, computing power, data and human capital) to be strong competitors in the generative AI space, or

(ii) AI partnerships may potentially negatively impact competition, in that they could be a tool through which, for example, big tech players could seek to expand their market positions into the provision of AI products and services or enable big tech players to tie up innovative FM developers in long term agreements to utilise computing power and/or cloud services exclusively from them.

The President of the French competition authority recently referred to these types of AI partnership as giving "the keys to the henhouse to the fox". It remains to be seen how regulators will actually approach the question, which will need to be considered on a case by case basis on the particular facts.

Most significantly in the UK so far, the CMA recently concluded that the partnership between Microsoft and Mistral AI to make Mistral's models available through the Microsoft Azure platform did not meet the jurisdictional thresholds for investigation under the very flexible UK merger control regime. (The CMA also has an open Phase 1 investigation into the Microsoft/Inflection AI deal, and the Microsoft/OpenAI, Amazon/Anthropic and Alphabet (Google)/Anthropic deals are in its pipeline for review).

The UK merger control regime has a particularly low control threshold when compared to other jurisdictions. In particular, the UK merger control regime can be engaged where there is an acquisition of 'material influence' (which is a lesser standard than the 'decisive influence' test in the EU). As a general rule, a shareholding of more than 25% is likely to be viewed as giving rise to material influence, and shareholdings of as low as 10-15% (with no board representation or other governance rights) might be viewed as conferring material influence, depending on the circumstances.

Indeed this was the key point considered in Microsoft/Mistral, as the CMA concluded that the partnership did not confer material influence over Mistral to Microsoft – even with the low control threshold. The CMA did, however, make some interesting comments about when the material influence threshold might be crossed, i.e. where there is a "dependency" relationship arising between two players. Such a relationship may, in particular, arise where the AI pure player, or foundation model developer, becomes dependent on the big tech player or cloud provider.

        The CMA has said that "dependency" might arise, for example:
    
By way of commitment to future collaboration and development opportunities;
By way of distribution agreements between the cloud service provider and the FM developer, where the agreement creates a dependency on the cloud services or allows the cloud services provider to influence the commercial policy of the FM developer; and / or
By way of provision of access to compute infrastructure if this "locks-in" the FM developer or compromises its ability to be commercially independent.

Since its Microsoft/Mistral decision, the CMA has opened a Phase 1 merger investigation into arrangements entered into between Microsoft and Inflection AI. Under those arrangements, Microsoft is reported to have hired certain former employees of Inflection AI and entered into a non-exclusive licensing deal that would allow Microsoft to use Inflection AI’s existing models. The deadline for the CMA's Phase 1 decision is 11 September 2024.

Similar themes have come into recent focus for the European Commission in its review of whether the Microsoft/OpenAI partnership might plausibly be subject to scrutiny under the EU Merger Regulation. The Commission concluded that it would not be, on the basis of the higher threshold for control in the EU. However, despite this, the Commission has said that it will still seek to scrutinise the partnership by exploring whether the exclusive use of Microsoft's cloud technology, Azure, by OpenAI could hamper competition as regards Microsoft's competitors. In short, where the Commission cannot deploy its merger control powers, it is seeking to look at the deal from an antitrust perspective. The outcome of this case is keenly awaited.

Subject to the question of "control", It is worth noting that the DMCC Act may become relevant to the requirement to notify M&A deals in the AI space.

First, the new 'no increment' test designed to catch 'killer acquisitions' set out in the DMCC Act may apply to deals in which large acquirers buy start-ups (for example new generative AI providers) – these provisions are expected to come into force in the Autumn, after the DMCC Act received Royal Assent following the pre-election wash-up (see our briefing here).
Second, the DMCC Act requires large tech firms that are designated with 'strategic market status' to report certain of its M&A activity to the CMA (see here).

Inputs to Generative AI

In relation to AI specifically, the CMA's initial review into FMs focussed on securing access to key inputs for development of FMs, including cloud services. The CMA noted that access to critical inputs could be restricted or used to shape the development of FM-related markets in the interests of incumbent firms.

        The key inputs highlighted by the CMA are:
    
Computing power, including specialised chips and cloud services;
Data, including training data and fine-tuning data;
Technical expertise, including both broadly and highly skilled workforce (e.g. research scientists and engineers); and
Capital and investment to fund the acquisition of the other inputs required.

It is notable that each of the main workstreams discussed in section 3 above (the CMA's review of Microsoft/Mistral, the Commission's review of Microsoft/OpenAI and the comments to date from competition authorities) are focussed on the importance of cloud services as an input to generative AI models.

Cloud services attract significant mention in discussions around competition in AI, perhaps because cloud services provision has been its own hot competition issue over the last few years. The CMA is currently carrying out a standalone market investigation into cloud services (which the CMA referred to in its update paper on FMs as a key next step in its work on AI). A key concern being explored under that market investigation is whether customers of cloud services are unlikely to switch between providers and, in fact, may find themselves contractually and/or practically "locked-in" to a particular provider.

Also noteworthy from the list of key inputs is that:

The CMA will examine the competitive landscape in AI accelerator chips, and the impact this has on the FM value chain, as part of its next phase of work on the FMs initial review. France and the US are also currently looking at antitrust investigations targeted at Nvidia, a significant chip manufacturer for AI.
As regards technical expertise and access to talent, another hot topic for competition authorities in all sectors centres around labour market practices. The practices of interest to competition authorities include wage fixing agreements, non-competes and non-poaching agreements. In markets where access to a talented workforce is an important parameter of competition, such as in AI markets, it is important to keep in mind competition risks in relation to labour market activities. A particularly stark example of enforcement in this area (though not in the context of AI) is that the U.S. Department of Justice recently conducted a number of criminal prosecutions against employers who allegedly either agreed with one another to fix employee wages or not to poach each other’s employees.

Cross-border Cooperation

Many authorities across the globe are independently considering 'access to inputs' issues in various different ways. The CMA's strategic update notes that it will be working closely with the ICN, OECD (the Competition Committee and separately the Committee on Consumer Protection), the G7 (AI Working Group), and the ICPEN to share insights, approaches, knowledge and expertise on AI.

The CMA has also recently signed a joint statement with the European Commission, US Federal Trade Commission and US Department of Justice setting out shared principles to support competition and protect consumers in the AI space. Those common principles are fair dealing, interoperability and choice. The statement affirms the agencies' shared view that AI (in particular powerful FMs) has the potential to spur positive innovation and growth, but also may pose some risks to competition and consumers which, if they materialise, will require action before they become entrenched or irreversible harms. The principles set out by earlier by the CMA in the context of its work on FMs (see section 2 above) still, however, remain relevant: the CMA has confirmed that those earlier principles continue to be developed and deployed.

UK National Security Considerations

M&A activity in the AI space is not only of interest to competition regulators. AI is one of the specified mandatory sectors under the UK's National Security & Investment Act ('NSIA'). This means that the UK's Investment Security Unit of the Cabinet Office ('ISU') must also be notified of (and give approval for) deals under the UK's NSIA where the target business carries out certain AI related activities.

The ISU's guidance notes that there are certain national security risks associated with AI, including the potential for actors to deploy AI applications for malicious and harmful uses. The notification regime can capture entities that do not necessarily identify as ‘AI companies' but which incorporate or develop AI as part of a wider approach to their sector or business.

However, in response to the previous UK Government's Call for Evidence on the NSIA, a number of respondents flagged that they would welcome clearer definitions for the AI sector on the basis that the current definition captures activities that do not present national security risks. The then-Government committed to launch a public consultation on updating the mandatory sectors this summer. That work is expected to continue with the newly appointed Labour government, although the precise timing and scope remain to be seen.

Key Take-aways for Businesses Active in the AI Space

A range of bodies, both within the UK and internationally, are focused on AI, competitive development of AI markets and enforcement in the AI space. In the competition law arena, given the evolving nature of the products and markets involved and the broad range of tools at the disposal of competition authorities, if would be prudent for market operators to exercise a degree of caution (both in their M&A strategies and their day-to-day commercial activities involving AI).

Key overarching points for AI-active businesses to consider include:

Whether their AI related activities are compliant with applicable competition laws. This might include, for example:
- Developing an antitrust compliance programme for AI related activities specifically.
- Considering interoperability between AI applications and models with other software and hardware.
- Ensuring that agreements to supply key inputs for AI, whether that be a rich data set or cloud computing services, are non-discriminatory and competition law compliant - particularly where those agreements are exclusive in nature.
Whether any proposed AI commercial endeavours with third parties would be compliant with applicable competition laws and/or may be subject to merger control scrutiny.