Regulating artificial intelligence: Between the EU Digital Markets Act and competition law

Abstract

The Digital Markets Act (DMA) entered into force in November 2022, with the aim of making digital markets fairer and more contestable.[1]  Designated gatekeepers providing core platform services have to comply with a list of “dos” and “don’ts” intended to, among other things, promote interoperability, avoid self-preferencing on the part of gatekeepers, and enable effective data portability.

However, questions on the applicability of the DMA to AI, which have been a point of debate since the Act’s entry into force,[2] have become more pressing with the growth of AI.  “AI” is not currently listed as a core platform service (in part because defining “AI” is not an easy task).  While some commentators have argued that the DMA already applies to AI, where AI forms part of an established core platform service, a recent consultation launched by the European Commission (EC) on how the DMA can support fair and contestable markets in the AI sector suggests there may be lacunae.  Further, the EC has recently launched two investigations (into Google and WhatsApp’s integration of AI features and products) for potential breaches of “traditional” competition law rules on the basis that these offer a more “holistic” framework than the DMA.[3]

So, does AI spell the premature end of the DMA? This article will argue that, while traditional competition law provides greater flexibility in tackling potential anticompetitive effects arising out of AI which may not fall neatly into core platform services, the DMA’s applicability to the AI sector should be tested to its fullest extent.

1.      Introduction: purpose of the DMA and its relationship to Art 101 and Art 102 TFEU

The DMA is an ex ante regulatory instrument which seeks to ensure the fairness and contestability of markets in which some of the largest digital companies operate.  Gatekeepers providing core platform services (CPSs) [4] such as social networks, operating systems, browsers and virtual assistants must comply with a series of obligations and prohibitions designed to promote interoperability and avoid self-preferencing.  At the time of writing, Alphabet, Amazon, Apple, Booking, ByteDance, Meta and Microsoft have been designated as gatekeepers in respect of one or more CPSs.

By contrast, Articles 101 and 102 of the Treaty on the Functioning of the European Union (TFEU) are “traditional” laws aimed at enforcing ex post the healthy functioning of the internal market by prohibiting anticompetitive agreements and abuses of dominant positions respectively.

The DMA’s relationship to Articles 101 and 102 is set out in the Act’s recitals.  Whereas Articles 101 and 102 “apply to the conduct of gatekeepers”, their scope is “limited to certain instances of market power, for example dominance on specific markets” and “enforcement occurs ex post” following an extensive investigation.[5]  Specifically, the DMA notes that competition law (in contrast to the DMA) “does not address effectively the challenges […] posed by the conduct of gatekeepers that are not necessarily dominant in competition-law terms”.[6] 

The DMA is portrayed as complementary to the enforcement of competition law and applies without prejudice to Articles 101 and 102 (and corresponding national competition law rules).[7]  However, whereas the latter concern “the protection of undistorted competition” on “any given market”, the DMA seeks to ensure that (digital) markets with gatekeepers “are and remain contestable and fair, independently from the actual, potential or presumed effects of the conduct of a given gatekeeper covered by this Regulation on competition on a given market”.[8]  Finally, the EC can conduct market investigations to identify whether the DMA needs to be amended to capture conduct by gatekeepers which is not currently covered by the DMA’s obligations, and can do so without prejudice to the EC’s ability to open proceedings under Articles 101 or 102 TFEU.[9]

2.      How might the DMA currently apply to AI?

When the DMA entered into force, it was part of a suite of measures the EC sought to introduce to regulate contemporary digital markets – together with the Digital Services Act and the EU AI Act.  Its entry into force in November 2022 also coincided with the release of OpenAI’s ChatGPT which familiarised the public with general purpose AI models and chatbots built on the same.

The question as to whether the DMA applied to AI technologies was therefore one that arose from the regime’s entry into force.  The EC’s primary response, as best encapsulated in a statement issued by the High-Level Group for the DMA in May 2024[10]  and in a Competition Policy Brief on Generative AI in September 2024,[11] was that:

(i) increased adoption of AI technologies has the potential to impact the contestability and fairness of digital markets;

(ii) this can lead to the entrenchment of existing gatekeepers, or the emergence of new ones; and

(iii) the DMA is engaged where either (a) a generative AI provider offers a CPS and meets the thresholds to be designated a gatekeeper or (b) a gatekeeper embeds AI into a designated CPS.

Accordingly, while “AI” is not listed as a CPS in the DMA, because AI is increasingly a feature embedded within other CPSs - rather than as a standalone service - it is arguable that the DMA already applies to the provision of AI-powered services.

This can be demonstrated by way of two examples: 

(a)    AI services falling within the definition of an existing CPS

The DMA provides that “virtual assistants” can be a CPS and that could be interpreted as encompassing end-user interfacing applications like ChatGPT or Claude. Virtual assistants are defined in the DMA as “software that can process demands, tasks or questions, including those based on audio, visual, written input, gestures or motions, and that, based on those demands, tasks or questions, provides access to other services or controls connected physical devices”,[12] a definition originally crafted to refer to services such as Amazon’s Alexa or Apple’s Siri. Accordingly, depending on whether it is necessary for chatbots like ChatGPT to provide other services or control physical devices, these may or may not fall within a CPS. If they do fall within the definition of a CPS, then the DMA’s provisions may bite.

(b)    A gatekeeper embeds AI into a designated CPS

The DMA prohibits gatekeepers from self-preferencing in ranking, indexing and crawling and requires the gatekeeper to apply transparent, fair and non-discriminatory conditions to any such ranking.[13] Commentators have argued that, where generative AI is integrated into a designated CPS like search, this prohibition could apply to AI-driven integrations that determine search prominence, such as Google’s Overview boxes now appearing at the top of Google Search.[14] The fact that AI-integrated functionalities would be caught by the DMA’s rules because they are embedded into an existing CPS also reflects comments made by the High-Level Group for the DMA on AI.[15]

3.      What are the DMA’s current limitations in respect of AI?

Some commentators have however pointed to the limitations of the DMA in its current form. In particular, they argue that, by not classifying “generative AI” as a CPS, providers such as OpenAI or Anthropic are unlikely to be designated as gatekeepers and therefore have an obligation to comply with the DMA.[16] This is on the basis that, in their view, while AI chatbots process “demands, tasks or questions”, their primary purpose is not to “provide access to other services” or “control connected physical devices” but rather to simply provide generative responses directly. Coupled with the understanding that “virtual assistants” were originally defined with physical devices like Amazon’s Alexa in mind, these commentors consider that ChatGPT and Claude fall outside the remit of the DMA.[17]   Similarly, in view of the fact that the DMA was not drafted with contemporary forms of AI in mind, its data restrictions are said to be poorly aligned with AI development stages (pre-training, fine-tuning, deployment), with the DMA kicking in - if at all - only once an AI service is effectively deployed within a CPS and thus arguably ignoring any concentration of data in pre-training stages.[18] 

In August 2025, the EC launched a public consultation as part of its obligation, under Article 53 of the DMA, to evaluate by 3 May 2026 (and every three years after then) whether the DMA has achieved its objectives.  The consultation sought, amongst other things, input on whether existing DMA obligations are sufficient to ensure a contestable and fair AI sector.[19] 

In January 2026, the EC published a summary of the responses it received.[20]  While a large number of respondents sought clarity on whether, and if so which, AI services fall within existing CPS categories, two positions reflecting the above debate emerged: the first submitting that new AI functionalities and services can be covered by the current list of CPSs and thus already fall within the DMA’s scope, and the second arguing that new standalone CPS categories (such as AI foundation models and chatbots) would need to be added in order for the DMA to cover AI.   

Similarly, in December 2025, a joint paper by members of the DMA Higher Level Group on AI mapped out the regulatory interplay related to AI issues.[21] The joint paper identified access to AI infrastructure and distribution and access to data as two interconnected areas where the DMA might  contribute to ensuring contestability and fairness ex ante, provided that consistent terminology relating to AI is adopted so as to address potential enforcement gaps.

Thus, the EC appears to recognise (i) the need for greater clarity regarding whether AI services fall within the DMA’s current scope and (ii) that regulatory gaps may exist between the DMA’s ex ante application and the ex post application of Articles 101 and 102.

4.      How has this been reflected in recent EC investigations?

As set out in Section 1 above, the DMA and “traditional” competition law rules are intended to be complementary to each other, with the DMA (according to its recitals) purportedly offering a more flexible approach to regulation on the basis of its ex ante interventions and presumption of a gatekeepers’ impact on the market without the need for extensive investigations into the effect of a dominant firm’s conduct.

However, recent EC investigations seem to suggest that the DMA – absent the clarifications sought by respondents to the EC's consultation – might be too rigid for the EC itself to carry out investigations into AI services even when these are offered by companies already designated as gatekeepers under the DMA.

By way of example:

On 4 December 2025, the EC opened formal competition law proceedings into Meta to assess whether its recent policy to exclude third party AI assistants from accessing and interacting with users on WhatsApp constituted an abuse of dominant position under Article 102. In February 2026 the EC notified Meta of its intentions to impose interim measures.[22]

Similarly, on 9 December 2025, the EC opened a competition law investigation to assess whether Google breached Article 102 in using the content of web publishers to provide its AI Overviews and AI Mode on search result pages, as well as content uploaded on YouTube, by (i) imposing unfair terms and conditions on publishers and content creators or (ii) granting itself privileged access to such content.[23] 

It is arguable that both investigations could have been pursued by the EC under the DMA’s provisions regarding self-preferencing, especially since Alphabet and Meta have already been designated as gatekeepers in respect of the relevant CPSs:  Google Search, YouTube and WhatsApp.  Nevertheless, EC Commissioner Ribera noted (in relation to the Google investigation) that while aspects “could have worked” under the DMA, it was “much more effective” to carry out the investigation under the traditional competition law abuse of dominance framework.[24]  This suggests that, in both cases, the EC considered the DMA to be either too inflexible to apply to Google and Meta’s conduct or insufficiently clear in its scope for the EC to confidently pursue its investigations as DMA cases.

Perhaps in an attempt to emphasise that the DMA is intended to be complementary and applies without prejudice to Article 102, in January 2026, the EC opened proceedings to assist Google in complying with its DMA obligations  to grant  third-party providers of online search engines access to anonymised ranking, query, click and view data held by Google Search on fair, reasonable and non-discriminatory terms, with a focus on (amongst other things) the eligibility of AI chatbot providers to access the data.[25]

5.      What can we expect next?

As recognised by the EC in its consultation summary and as exemplified by the recent EC investigations above, the DMA’s application to AI needs to be clarified. While the EC is due to publish its review report pursuant to Article 53 of the DMA by 3 May 2026, it remains to be seen whether the EC launches any further investigations under the DMA or under its “traditional” competition law rules.

The EC could of course test the applicability of the DMA to AI services itself, by being bolder with its investigations – especially in relation to services (like chatbots) which arguably already fall within the definition of “virtual assistants”.

Nonetheless, while clarifications on the applicability of the DMA will no doubt be welcome, it seems clear that both the DMA and competition law rules will continue to operate in parallel, or indeed in tandem, with regards to existing and novel AI services.