⚡ Quick Summary
Published by the Centre on Regulation in Europe (CERRE), this issue paper by Friso Bostoen and Jan Krämer, dated November 2024, asks whether the EU's digital acquis is prepared for AI agents. It adopts the UK Department for Science, Innovation and Technology definition of AI agents as "autonomous AI systems that perform multiple sequential steps" and distinguishes them from the generative models they rely on. The scope is deliberately narrow: AI agents on mobile devices and wearables, not general competition issues related to AI.
The paper sketches the AI agent value chain — AI models, mobile AI chips, cloud computing resources, OS integration and application integration — and argues that firms already controlling bottlenecks such as operating systems, search engines and browsers are best placed to win the AI agent market. It lists contestability concerns including vertical foreclosure, discriminatory access conditions, exclusivity agreements, pre-installation and defaults, and limited data portability.
The core of the paper reviews the Platform-to-Business Regulation, the Digital Markets Act, the Data Act, the AI Act, the Digital Services Act, competition law under Article 102 TFEU and the Net Neutrality Regulation. It concludes that AI agents fit awkwardly with established categories such as online platforms, search engines and virtual assistants, leaving their categorisation and the applicable obligations uncertain.
🧩 What’s Covered
- Introduction: frames AI as a transformative shift comparable to the analog-to-digital and desktop-to-mobile transitions, and asks whether the EU legal landscape advances contestability for AI-enabled services and devices, citing Deutsche Telekom's vision of an "app-less" AI phone.
- AI agents and current industry partnerships: adopts the UK DSIT definition of AI agents plus comparable definitions from IBM, Microsoft and OpenAI; distinguishes agents from foundation models; sets out the four autonomy characteristics from Chan et al. (2023) — underspecification, directness of impact, goal-directness and long-term planning; and maps partnerships such as Apple with OpenAI and Samsung with Google.
- Required bottleneck inputs: describes the value chain of AI models, mobile AI chips, cloud computing resources, OS integration and application integration, explaining why on-device chips, cloud access and deep access to device hardware and software matter for agents.
- Contestability concerns: lists vertical foreclosure, discriminatory and unfair access conditions, exclusivity agreements, advantages from pre-installation and defaults, and superior data access, and discusses mixing-and-matching and vertical interoperability.
- Categorisation: tests AI agents against the P2B Regulation, DMA, Data Act, AI Act, DSA, competition law and the Net Neutrality Regulation, including DMA gatekeeper criteria such as 45 million monthly active end-users, 10,000 yearly active business users and €7.5 billion EU turnover.
- Obligations: covers DMA uninstallation (Article 6(3)), virtual assistant choice screens, interoperability (Article 6(7)), competition enforcement, transparency duties across the P2B Regulation, DSA and AI Act, and data access under the Data Act, DMA and GDPR.
- Conclusions and open issues: finds the framework fragmented and notes coordination through the High-Level Group for the Digital Markets Act.
💡 Why it matters?
For regulators, competition authorities and compliance teams, the paper gives a structured first reading of where AI agents land in EU digital regulation and why the answer is often unclear. It identifies the obligations that could bite once an agent is designated — uninstallability, choice screens, interoperability and data portability — and raises practical questions such as whether an OS provider could avoid a choice screen by pre-installing a partner's agent. It also maps where competition law continues alongside the DMA, particularly for cross-level practices such as cloud–model partnerships and exclusivity arrangements.
❓ What’s Missing
The paper offers a conceptual and legal assessment rather than a settled answer: it repeatedly notes that categorisation of AI agents under the P2B Regulation, DMA, DSA and Data Act is unresolved and depends on interpretation, and it provides no draft text for a new core platform service category. There is no quantitative market analysis or empirical measurement of the gatekeeper thresholds. Coverage is limited to mobile devices and wearables in the EU, so cloud-only agents, non-EU regimes and sector-specific rules fall outside. Much of the analysis is prospective, and the authors note that a general, highly autonomous AI system does not yet exist.
👥 Best For
Best for EU digital-policy analysts, competition lawyers and regulatory affairs teams tracking how the DMA, AI Act and Data Act may apply to agentic systems, and for platform compliance teams assessing uninstallation, default-choice and interoperability duties. Also useful for researchers wanting a conceptual map of the AI agent value chain and its bottlenecks.
📄 Source Details
AI Agents and Ecosystems Contestability is an Issue Paper published by the Centre on Regulation in Europe (CERRE), written by Friso Bostoen (Tilburg University) and Jan Krämer (University of Passau; CERRE Academic Co-Director), dated November 2024. It runs to 30 pages including table of contents and references, with copyright held by CERRE and contact details info@cerre.eu and www.cerre.eu. No URL to the document itself is printed; the full text of all 30 pages was available.
