EU AI Act meets AI Agents

This blog highlights key arguments from our recent Tech Policy Press article, “The EU AI Act is Not Ready for Agents,” on how the relationship between the EU AI Act and AI agents is evolving and the governance challenges posed by increasingly autonomous AI systems. Read the original academic paper and the shorter Tech Policy Press article built upon it.

AI agents, systems that independently pursue complex goals with limited human oversight, have entered the mainstream. They are now widely used to produce software, conduct business activities, and automate everyday personal tasks. But they also introduce unique risks.

Why AI Agents Challenge the EU AI Act

In a new piece published in Tech Policy Press, The Future Society team members Kathrin Gardhouse and Amin Oueslati examine how effectively the EU AI Act—the most comprehensive AI regulation to date—governs AI agents. Our research suggests that the EU AI Act applies to agents in principle, but falls short at the operative compliance layer, as we demonstrate across five governance challenges: performance, misuse, privacy, equity, and oversight. Across each, existing regulatory assumptions begin to strain.

AI Agents and Emerging Governance Risks

Recent incidents underscore these challenges: from agents causing major infrastructure disruptions to cases where they are manipulated into leaking sensitive data. AI agents behave differently from earlier AI models: they operate over extended periods, adapt through interaction, and take actions in the world that may be difficult—or impossible—to reverse.

The Tech Policy Press piece draws on our broader academic paper, which develops these arguments in detail and can be accessed here.

How the EU AI Act Should Account for AI Agents

We do not challenge the Act’s underlying logic, but argue that its operationalisation must better account for AI agents. Upcoming technical standards for high-risk systems offer one such opportunity, while guidance from the AI Office could help clarify how obligations for general-purpose AI models apply in agentic deployment contexts.

For more on agents under the EU AI Act, read our report “Ahead of the Curve: Governing AI Agents under the EU AI Act”, which was the first comprehensive analysis of regulation of AI agents in the European Union.