Series
Blogs
Series
Blogs
Authors: Rachel Malloch, Daniel Green, Tara Rudra
As AI becomes increasingly embedded in our daily lives, the antitrust debate around AI is evolving. AI offers huge potential to deliver consumer welfare gains – better services, greater innovation and lower prices. But, as focus shifts from upstream foundation models to downstream AI-powered applications, competition authorities have become increasingly alive to antitrust and consumer risk: algorithmic collusion; exclusionary conduct by incumbents; and broader fairness issues (whether assessed as exploitative abuses or consumer harms).
Competition authorities are meanwhile being asked to perform a dual role – acting as enforcers and enablers of competition – tasked not only with preventing harms but with fostering fair and competitive AI-driven markets. In this post, we look at the core antitrust issues emerging in AI and how competition authorities are choosing between the enforcement tools available to them.
Mainstream access to AI tools, brought about by the release of ChatGPT in late 2022, prompted an initial wave of scrutiny focused on foundation models. This began with the use of “soft” tools. The CMA, for example, launched its initial review of AI foundation models in May 2023. In July 2024, the Commission, CMA, DOJ and FTC issued a joint statement on competition in generative AI foundation models and AI products. A key potential risk area identified through that work – the use of partnerships and investments by tech incumbents in foundation models – ultimately resulted in detailed review of multiple AI partnerships under both UK and EU merger control rules. These reviews clarified how merger control rules may apply to partnership and investment structures in the AI sector, but all ultimately resulted in unconditional clearance.
More recently, as agentic AI and quantum computing move from theory to reality, authorities are again focusing on initial fact-finding missions. The CMA published research and guidance on agentic AI in March 2026, setting out how consumer protection law applies to autonomous AI agents and flagging risks around dark patterns, reliability and algorithmic bias. Also in March 2026, the Italian authority (AGCM) launched a forward-looking sector inquiry into quantum computing, with a focus on barriers to entry and vendor lock-in.
The French competition authority (FCA) has been particularly active. Its work on AI-adjacent issues began with an opinion on cloud computing (June 2023), which examined the competitive dynamics of a key input market for AI. It has since issued a series of opinions tracking the AI value chain: an opinion on generative AI (June 2024) examining the upstream market for foundation models; a study on the energy and environmental impact of AI (December 2025); and, earlier this month, an opinion on AI agents examining the competitive dynamics of the downstream market, including the rise of agentic commerce. In its latest opinion, the FCA recommends as a matter of priority to “leverage the existing regulatory framework” – antitrust, the DMA and the AI Act – rather than rushing to legislate anew, in a context where “the time needed to develop new standards is difficult to reconcile with the pace of innovation in this field”.
This deliberately careful approach allows enforcers to upskill on nascent technology, while identifying and raising awareness of potential antitrust risks – promoting compliance and effective competition at an early and uncertain stage without inadvertently stifling innovation.
As understanding of AI tools and markets matures, and adoption becomes more entrenched, authorities are also signalling a willingness to consider tougher enforcement, drawing on traditional antitrust tools (as well as newer digital regimes – see further below). Three key trends are worth highlighting.
AI-powered pricing tools, which risk facilitating collusion or aligning pricing conduct through the exchange of commercially sensitive information, remain an enforcement priority. This risk is not new (and has been well trailed outside AI models) and was examined by the FCA and the German Bundeskartellamt as early as 2019. Attributing anticompetitive conduct to a specific undertaking is inherently difficult; a challenge only amplified by the rise of agentic AI. As a practical matter, authorities have been keen to insist that using an AI intermediary does not insulate a company from liability and that traditional antitrust rules apply (see our previous post on agentic AI for more). Enforcement in this area is ongoing in the US and UK, with several algorithmic pricing cases in the pipeline in the EU (see our post on this for more).
The position of Big Tech as potential gatekeepers of digital ecosystems has given rise to concerns that incumbents could unfairly leverage their existing positions to their advantage in AI. The Commission’s ongoing investigation into Google’s AI and data practices, opened in December 2025, is examining whether Google is granting itself privileged access to publisher and YouTube content, placing rival AI models at a disadvantage. In Meta/WhatsApp, the Commission imposed interim measures in June 2026 – only the second time the Commission has used this tool since 2004 – to restore free access to WhatsApp for rival AI assistants. Its decision to impose interim measures, given the circumstances of the case and the alleged theory of harm, is unprecedented. The principles underpinning exclusionary theories of harm developed in earlier digital cases – tying, leveraging, self-preferencing – are being deployed afresh to address potential AI-related harms.
Exploitative abuses, historically not a focus for enforcers outside specific market failings (e.g. excessive pricing in pharmaceuticals), have increased in use in recent years. Alongside the self-preferencing theory of harm, the Commission’s Google AI investigation examines whether Google imposed unfair terms and conditions on publishers and content creators by using their content in its AI Overviews without compensation or a right of refusal. This builds on the Commission's willingness to pursue exploitative abuses in earlier digital cases, as well as statements from Commission officials signalling a “pivot” towards enforcing against exploitative conduct.
What is driving this? Principally, we see the Commission using exploitative tools to address broader issues around fairness in a world of digital ecosystems – including, unusually, in respect of B2B conduct. This trend, however, comes with questions about the appropriate legal test to determine the boundaries of exploitative abuses – with challenges against Commission decisions, including Meta’s action for annulment against the Facebook Marketplace decision.
This appetite to address issues around fairness is equally seen at the NCA level, including through a consumer law lens. The Italian AGCM, for example, closed investigations into DeepSeek, Mistral AI and NOVA AI in April 2026 via commitments aimed at improving transparency around AI systems and, in particular, the risk of hallucinations.
Concern around whether the EU’s Digital Markets Act is fit to address AI-related antitrust issues has grown alongside the rapid adoption of generative - and now agentic - AI. Certainly, the architects of the regime did not anticipate the transformative impact of AI during its design. For now, however, the Commission’s official view is that it can accommodate AI-related issues within the current DMA framework.
The Commission's first formal review of the DMA, published in April, concluded that the DMA remains “fit for purpose” and requires no legislative amendment – but identified AI and cloud computing as priority enforcement areas going forward. It also points to concrete steps the Commission is already taking to address both fairness and contestability issues in AI. These include regulatory dialogue with gatekeepers on AI-related issues and ongoing monitoring of the deployment of AI tools within designated core platform services.
Most significantly, the Google specification proceedings under Articles 6(7) and 6(11) directly address potential leveraging concerns in the context of Google’s unique advantages across the AI stack. Earlier this month, those proceedings concluded with the adoption of two binding decisions requiring Google to (i) give rival AI assistants equally effective access to Android hardware and software features used by Gemini, and (ii) to share anonymised search ranking, query, click and view data with competing search engines and AI chatbots on FRAND terms. Google has until January 2027 to begin sharing search data and until July 2027 to implement the Android interoperability measures. Last week, the Commission fined Google EUR 890m for non-compliance with the DMA in respect of self-preferencing on Search (as well as steering rules in respect of Google Play) with the Commission referring to ongoing dialogue with Google on how the principles in the decision apply to AI Overviews and AI Mode.
Importantly, the Commission's review keeps the door open to possible designation of AI services within the existing "virtual assistants" core platform service category – originally designed with the likes of Siri and Alexa in mind – or the creation of an entirely new core platform service category for AI.
While the Commission maintains the DMA rules need not change at present, it is clear that abuse of dominance is, for now, its preferred tool to address potential AI antitrust concerns. When announcing the Google AI investigation in December 2025, Commissioner Ribera acknowledged that while aspects “could have worked” under the DMA, it was “much more effective” to pursue the case under the traditional abuse of dominance framework, because it offered a more “holistic” approach to conduct that does not fit neatly within the DMA’s existing obligations. This is notable given that a key driver of the DMA was the recognition that ex post antitrust enforcement was too slow, too cumbersome and too late.
In contrast, the UK’s digital markets regime – under the Digital Markets, Competition and Consumers Act (DMCC) – is more flexible by design given that, unlike the EU framework, the UK regime applies bespoke rules in the form of conduct requirements to firms designated as having strategic market status (SMS).
Within its current SMS designations, the CMA is tackling AI-related concerns as they arise within the designated digital activities. It has imposed conduct requirements under the DMCC (described by the CMA as a “world first”) to address fairness issues around use of content in AI Overviews in Google Search – the same concern which is, in contrast, being examined under abuse of dominance rules by the Commission. This preference for the DMCC over traditional abuse of dominance enforcement is clear and seems set to continue. The CMA's prior docket of tech abuse of dominance cases – notably its Google Play Store and Apple App Store investigations – were closed on the basis the CMA’s identified concerns could be considered “in a more timely, holistic and flexible manner” under the DMCC instead.
The picture that emerges is one of pragmatic adaptation rather than regulatory revolution. Across jurisdictions, competition authorities are reaching the same conclusion: existing tools can, for now, flex to meet the challenges posed by AI. The FCA’s latest opinion captures this neatly, recommending that enforcers mobilise frameworks already in place before reaching for new legislation.
That pragmatism comes with a tension. The Commission’s preference for abuse of dominance enforcement over the DMA is a tacit acknowledgment that speed and specificity – the very qualities the DMA was designed to deliver – can come at the cost of flexibility. The UK’s DMCC regime, by contrast, appears for now to be the CMA’s tool of choice to address AI-related issues.
What is beyond doubt is the pace at which this landscape is evolving. Foundation models, generative AI, agentic commerce – each wave has arrived faster than regulators (and market participants) anticipated. Competition authorities are watching closely, and the enforcement pipeline is growing. The regulatory toolkit may be familiar, but the targets are new, and enforcers are not waiting for the rulebook to be rewritten before acting.
27 July 2026