EU says agents fall under AI Act

The European Commission has quietly told companies that autonomous artificial intelligence agents are covered by the European Union’s Artificial Intelligence Act. (ai-act-service-desk.ec.europa.eu) The Commission’s AI Act Service Desk added a new frequently asked question on “How are AI agents addressed within the AI Act?” by early April 2026, and the International Association of Privacy Professionals reported the update on April 2, 2026. (ai-act-service-desk.ec.europa.eu) (iapp.org) The Act does not create a separate legal category called “AI agent.” Instead, the law already defines an artificial intelligence system as software that operates with varying levels of autonomy and can generate predictions, content, recommendations, or decisions that affect real or virtual environments. (eur-lex.europa.eu) (artificialintelligenceact.eu) That matters because a company using an agent is usually a “deployer” under Article 3, while the company building or branding it is usually the “provider.” Those roles carry different duties under the Act. (artificialintelligenceact.eu) (nicfab.eu) If an agent is used in a high-risk setting, the deployer must follow the provider’s instructions, assign human oversight, manage input data, monitor operation, keep logs for at least six months, and report serious risks or incidents. (ai-act-service-desk.ec.europa.eu) (eur-lex.europa.eu) If an agent talks directly to people or generates synthetic audio, images, video, or text, separate transparency rules also apply. Those Article 50 obligations start applying on August 2, 2026. (ai-act-service-desk.ec.europa.eu) (springlex.eu) The timing is important because the law is phasing in. Prohibited practices and artificial intelligence literacy duties started on February 2, 2025, general-purpose model rules started on August 2, 2025, and most high-risk system rules and broader enforcement start on August 2, 2026. (ai-act-service-desk.ec.europa.eu) (digital-strategy.ec.europa.eu) Europe’s enforcement design is also more centralized than many companies expected. The European Artificial Intelligence Office was formally established on February 21, 2024, inside the Commission, and it oversees general-purpose model obligations at European Union level while national authorities handle other parts of the Act. (eur-lex.europa.eu) (digital-strategy.ec.europa.eu) (epthinktank.eu) The Commission has said its guidance is not legally binding, but it also says the documents reflect its interpretation and will guide enforcement. For companies shipping agents into Europe, that makes the new frequently asked question more than a glossary note. (digital-strategy.ec.europa.eu) (ai-act-service-desk.ec.europa.eu)