1、June 2025Governing AI AgentsUnder the EU AI ActAmin Oueslati andRobin Staes-PoletAhead of the Curve Ahead of the Curve:Governing AI Agents Under the EU AI Act Ahead of the Curve:Governing AI Agents under the EU AI Act Contact:infothefuturesociety.org Cite as:Amin Oueslati and Robin Staes-Polet,“Ahea

2、d of the Curve:Governing AI Agents under the EU AI Act The Future Society,June 2025.2025 by The Future Society.Photography by Christian Lue.This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License .Ahead of the Curve:Governing AI Agents Under the EU AI Act T

3、able of Contents Executive Summary 1 Glossary 5 1.State of AI Agents 7 1.1 AI Agent Trends 8 1.2 Characterising AI Agents 11 1.3 Evolving Landscape 12 2.Risks from AI Agents 14 2.1 Risk Mechanisms 14 2.2 Chapter V Agents and Systemic Risks of GPAI Models 15 2.3 Chapter III Agents and Domain-Specific

4、 High-Risk Systems 16 3.Applicability of the EU AI Act 18 3.1 AI Agents as Systems 19 3.2 AI Agents as GPAI Systems 21 3.3 AI Agents Built on General-Purpose AI Models with Systemic Risk 22 3.4 AI Agents as High-Risk AI Systems 22 3.5 Enforcement Authority 23 3.6 Open Questions and Working Assumptio

5、ns 23 3.7 Allocation of Obligations Along the Value Chain 25 3.7.1 Illustrative Example:Risk Identification Along the Value Chain)26 4.Governance Along the Value Chain 28 4.1 Agent Risk Assessments 30 4.1.1 Risk Identification 30 4.1.2 Risk Evaluation 31 4.1.3 Challenges and Future Directions 32 4.2

6、 Transparency Tools 32 4.2.1 Agent Identifiers 33 4.2.2 Real-Time Monitoring 34 4.2.3 Activity Logs 35 4.2.4 Acceptable Use Policies 36 4.2.5 Challenges and Future Directions 37 4.3 Technical Deployment Controls 38 4.3.1 Real-Time Action Refusals 38 4.3.2 Emergency Shutdowns 39 4.3.3 Challenges and