EU treats AI agents as regulated

The European Commission’s AI Act help desk now says artificial intelligence agents are covered by the law when they meet the Act’s definition of an artificial intelligence system. (ai-act-service-desk.ec.europa.eu) The clarification appeared in the Commission’s public frequently asked questions on the AI Act Service Desk, the portal the Commission launched on October 8, 2025 to answer compliance questions with the European Union’s Artificial Intelligence Office. (ai-act-service-desk.ec.europa.eu, digital-strategy.ec.europa.eu) An artificial intelligence agent is software that does more than chat back: it can plan steps, call tools, and take actions in other systems with limited human input. The Commission’s scope page says the Act applies to anyone placing artificial intelligence systems on the European Union market, putting them into service, importing them, distributing them, or using them in the bloc. (ai-act-service-desk.ec.europa.eu) That moves the argument from “are agents covered at all?” to “which duties apply to this product?” The Artificial Intelligence Act uses a risk-based structure, with bans for some uses, separate duties for general-purpose models, and heavier obligations for high-risk systems. (digital-strategy.ec.europa.eu) The dates are no longer theoretical. The Commission’s implementation timeline says general-purpose model rules started applying on August 2, 2025, and that most of the Act, including transparency rules in Article 50 and enforcement at national and European Union level, starts on August 2, 2026. (ai-act-service-desk.ec.europa.eu) That matters for companies building assistants that can browse, buy, route tickets, approve steps, or trigger workflows inside banks, hospitals, employers, and public agencies. If an agent lands in a high-risk use case listed by the law, the provider and the deployer can face documentation, oversight, logging, and risk-management duties instead of treating it like a simple chatbot. (digital-strategy.ec.europa.eu, ai-act-service-desk.ec.europa.eu) The Commission has been building tools around that rollout. Its Single Information Platform includes a compliance checker, an article-by-article explorer, and a service desk form for questions sent to experts working with the Artificial Intelligence Office. (digital-strategy.ec.europa.eu, ai-act-service-desk.ec.europa.eu) Vendors are already selling around the new compliance burden. On April 14, 2026, Cisco and Belgium-based ML6 said they signed a memorandum of understanding to accelerate secure artificial intelligence in Europe and to help customers prepare for compliance with the European Union Artificial Intelligence Act. (ml6.eu) ML6 said it is joining Cisco’s Artificial Intelligence Defense Design Partner Program, and Cisco said its tooling is meant to assess model vulnerabilities and add real-time guardrails across cloud, hybrid, and on-premises deployments. (ml6.eu) The Commission’s message is narrower than some of the marketing around it: “agent” is not a special legal category with its own chapter. It is a product label, and the legal test remains whether the system fits the Act’s definition and where it sits in the law’s risk tiers. (ai-act-service-desk.ec.europa.eu, digital-strategy.ec.europa.eu) By August 2, 2026, European companies will not be debating whether autonomous software belongs inside the Artificial Intelligence Act. They will be mapping each agent to the rulebook that already exists. (ai-act-service-desk.ec.europa.eu, ai-act-service-desk.ec.europa.eu)