EU: agents fall under AI Act
European agencies have clarified that AI agents are subject to the EU AI Act and analysts argue this pushes organisations toward centralised governance rather than a patchwork of local controls. The Commission’s follow‑ups and related commentary stress coordinated implementation and practical compliance tooling for agents across markets. ( )
The European Commission’s new guidance says artificial intelligence agents are not a separate legal carveout under European Union law and can fall inside the Artificial Intelligence Act. (ai-act-service-desk.ec.europa.eu) That matters because the Act regulates “artificial intelligence systems” by function and risk, not by product label, under Regulation (EU) 2024/1689. The law entered into force on 1 August 2024 and applies in stages across 2025, 2026 and 2027. (digital-strategy.ec.europa.eu, ai-act-service-desk.ec.europa.eu) In practice, an agent is software that can take inputs, make inferences and act across tools or services with some autonomy. If that system is used in a high-risk setting, or triggers transparency duties, the existing Artificial Intelligence Act rules apply to the provider, deployer or other actor in the value chain. (eur-lex.europa.eu, digital-strategy.ec.europa.eu) The Commission has spent the past year building a more centralised compliance architecture around that idea. Since 2 August 2025, the European Artificial Intelligence Board has been operational, the Scientific Panel’s rules have been adopted, and Member States have had to designate national authorities for enforcement. (interoperable-europe.ec.europa.eu) Brussels is also adding common tools instead of leaving companies to stitch together local answers. The Artificial Intelligence Act Service Desk now hosts a Single Information Platform, an Artificial Intelligence Act Explorer and a beta Compliance Checker for organisations trying to map their obligations. (ai-act-service-desk.ec.europa.eu, ai-act-service-desk.ec.europa.eu, ai-act-service-desk.ec.europa.eu) The Commission said on 4 December 2025 that it was preparing 2026 guidance on high-risk classification, transparency, serious-incident reporting, value-chain responsibilities, substantial modification and the Act’s overlap with other European Union laws. Those are the operational questions companies face when an agent calls models, tools and databases across borders. (digital-strategy.ec.europa.eu) A similar pattern already exists for general-purpose artificial intelligence models. The Commission published its Code of Practice on 10 July 2025, and said providers that sign it can use the code as a voluntary way to demonstrate compliance with obligations on transparency, copyright, and, for the most advanced models, safety and security. (digital-strategy.ec.europa.eu) The Commission’s Artificial Intelligence Continent Action Plan, published on 9 April 2025, paired that regulatory buildout with a call for coordinated action at European Union, national and local level. The plan said Europe’s advantage includes “one single set of safety rules across the EU,” language that fits the push toward shared governance for agent-based systems. (interoperable-europe.ec.europa.eu, interoperable-europe.ec.europa.eu) The upshot is that companies building agents for Europe are being told to classify the system they actually deploy, document who does what in the value chain, and use common European Union guidance rather than treating agents as an unregulated category. The law did not create a special box for agents; Brussels is making clear they still have to fit inside the one it already built. (ai-act-service-desk.ec.europa.eu, digital-strategy.ec.europa.eu)
