Is the DMA Ready for Agentic AI?

The future of artificial intelligence is agentic. The first signs of AI agents—systems which are more proactive and autonomous than current tools like ChatGPT—are already starting to emerge. These systems exhibit proactive, autonomous behaviour based on a user-set prompt or goal, adapt through learning, integrate with external services and interact with their environment. This may radically transform how consumers interact with their devices, with implications for competition in markets ranging from mobile operating systems to search and e-commerce.

In this CERRE report, Friso Bostoen and Jan Krämer examine whether the EU’s Digital Markets Act (DMA) is equipped for the advent of agentic AI. They identify two potential competition risks:

1. Foreclosure of AI agents – foreclosure by an existing operating system provider with its pre-installed own AI agent or another AI agent
2. Foreclosure by AI agents – where powerful agents themselves become gatekeepers and steer demand anticompetitively.

The authors find that the DMA is surprisingly well-suited to address the near-term risks of foreclosure of AI agents, particularly through its obligations on mobile operating systems—already designated as gatekeepers. Existing provisions on interoperability, default status, and uninstallation apply directly to AI agents when embedded in core platform services (CPSs) like OSs, search engines or virtual assistants. The report recommends further strengthening the regime by extending click and query data-sharing obligations to virtual assistants and including them in rules that prevent forced integration of ancillary services.

Looking further ahead, the report analyses the risk that AI agents may themselves become gatekeepers. It concludes that the DMA’s “virtual assistant” CPS category—while originally designed with voice assistants in mind—can plausibly accommodate AI agents. Some legal uncertainties remain, especially regarding the identification of business users and how well the current definition of virtual assistants reflects more autonomous AI capabilities. The authors recommend clarifying these.

In both scenarios, many of the DMA’s obligations—on interoperability, default settings, ranking neutrality, and data use—remain relevant.

The authors argue that agentic AI could, in the short term, be designated as virtual assistants under the DAM without disrupting legal certainty. In the mid to long-term, legislative change provides a more legally certain way to capture the future iterations of AI agents.

This report fills an important gap in the policy debate. While regulators have focused on foundation models, the agentic layer of AI remains largely unexamined. As the market evolves, the DMA could serve as a powerful tool to safeguard contestability—if adapted thoughtfully and with foresight.

The report was presented and debated on 3 July 2025, during the CERRE public webinar “AI and the Future of Competition“.