The Act has no separate category for agents. An agent is an AI system, usually built on a general-purpose model, and what it is used for decides what it owes.
This guide walks through how Regulation (EU) 2024/1689 applies to agents that plan and act: how to find an agent’s risk class, which duties fall on the company that builds it and which on the company that uses it, the application dates as amended in 2026, and the agent-specific points — autonomy, tools, oversight and logs. It explains the law; it is not legal advice, and whether a duty applies depends on your system and your role.
Updated · 7 sources
The EU AI Act for AI agents works through its general definitions: the Commission’s guidance says agents are not a separate category, because the definitions of an AI system and a general-purpose AI model already cover them. An agent’s duties follow from its intended purpose — prohibited, high-risk, subject to transparency rules, or none of these — and from whether you build it (provider) or use it under your authority (deployer). Transparency rules apply from 2 August 2026; high-risk rules from 2 December 2027 for Annex III uses.
An AI system under Article 3(1) is machine-based, operates with some autonomy, and infers from its input how to produce outputs such as predictions, content, recommendations or decisions that can influence physical or virtual environments. An agent fits that definition. Its class depends on what it is for — so the same model can sit in different classes in different agents.
What is the agent intended to do?
Profiling overrides the exemption. An Annex III system that profiles natural persons is always high-risk. A provider relying on Article 6(3) must document the assessment before release and register the system.
Purpose, not technology. An agent that summarises applications for a recruiter and one that ranks them sit in the same Annex III area; whether either is high-risk turns on what it does to the decision.
The provider develops an AI system and places it on the market or puts it into service under its own name. The deployer uses it under its own authority. A company that configures a general-purpose agent for a high-risk purpose can become its provider.
Develops the agent, or has it developed, and places it on the market or into service under its own name or trademark.
Uses the agent under its own authority in a professional activity — the bank, insurer or employer putting it to work.
The requirements in Chapter III were written for AI systems generally. For an agent — which chooses tools, takes several steps and changes records — some of them bite harder. The left column is the duty; the right is what it means for an agent.
The Act entered into force on 1 August 2024 and applies in stages. An amending regulation — Regulation (EU) 2026/1744, the AI omnibus, proposed by the Commission on 19 November 2025, agreed on 7 May 2026, published in the Official Journal on 24 July 2026 and in force since 27 July 2026 — moved the high-risk dates, added a ninth prohibition and softened the AI literacy duty.
Check the dates against the source. These are the dates in Articles 113 and 111 as amended by Regulation (EU) 2026/1744, and on the Commission’s AI Act page. Before the amendment, Annex III duties were due on 2 August 2026 and Annex I duties on 2 August 2027.
The Commission’s service desk FAQ, which it describes as preliminary, singles out a few points for agents. The scale below shows how the same kind of agent collects duties as its purpose moves toward decisions about people.
The FAQ flags harmful manipulation and exploiting vulnerabilities as the prohibitions most relevant to agents; safeguards belong in the design.
Lifetime event logs must let risks and modifications be traced — for an agent, the plan, tool calls and results.
Overseers must be able to interrupt the system through a stop procedure that leaves it in a safe state.
People must be able to disregard, override or reverse an output — for an agent, undo what it changed.
People are told they are interacting with AI unless it is obvious from the context.
For the model inside the agent, autonomy and tool use are among the criteria for a systemic-risk designation.
Which obligations apply follows from what your agent does and your role in it; your policies, risk assessments and people own the answers. The platform supplies controls and the evidence those answers rest on.
| Duty | Article | What the platform supplies |
|---|---|---|
| Keep records automatically | Art. 12 | Append-only, hash-chained audit trail with a verify call; a kept record of every agent run |
| Human oversight | Art. 14 | Approve, reject or edit on the tools you choose; review stages; cancel any run |
| Help users interpret outputs | Art. 13 | Each extracted value carries its source page, words and confidence |
| Accuracy and robustness | Art. 15 | Evaluation gates on every agent publish; calibrated confidence per field; drift monitoring |
| Risk management across the lifecycle | Art. 9 | Evidence for each step, alongside your own risk assessment and inventory |
| Control of change | Art. 11, Annex IV | Versioned configuration with diff, rollback and freeze |
Read the full control set in AI governance on Botminds and how people stay in charge in human in the loop. The Academy lesson on audit and compliance shows the audit trail in use.
Yes. Under the EU AI Act for AI agents, the Commission’s guidance says agents are not a separate category: the definitions of an AI system (Article 3(1)) and a general-purpose AI model (Article 3(63)) cover them. An agent’s duties follow from its intended purpose and from whether you are its provider or deployer.
Only when their intended purpose falls under Article 6 — mainly the Annex III areas such as employment, credit scoring of natural persons, education and essential services — and no Article 6(3) exemption applies. An agent that profiles natural persons in an Annex III area is always high-risk.
After the 2026 amendment: 2 December 2027 for Annex III systems and 2 August 2028 for AI in products covered by Annex I. Prohibitions have applied since 2 February 2025 and transparency rules from 2 August 2026.
Partly. Regulation (EU) 2026/1744, in force since 27 July 2026, moved the high-risk dates from 2 August 2026 to 2 December 2027 (Annex III) and from 2 August 2027 to 2 August 2028 (Annex I). The transparency rules kept their August 2026 date, and the amendment added a prohibition that applies from 2 December 2026.
Providers must design systems that interact with people so those people are told they are dealing with AI, unless it is obvious. Systems that generate synthetic audio, images, video or text must mark outputs in a machine-readable way. Deployers must disclose deepfakes.
The company that develops the agent and puts it into service under its own name is the provider of that AI system; the model maker remains the provider of the general-purpose model. A deployer that redirects an agent to a high-risk purpose becomes its provider under Article 25.
For high-risk systems, providers and deployers keep the automatically generated logs under their control for a period appropriate to the purpose, of at least six months unless other EU or national law says otherwise (Articles 19 and 26(6)).
No platform can do that alone; compliance depends on your system, your role and your processes. Botminds supplies controls and evidence — an audit trail, approvals, sourced outputs, evaluation gates and versioned change — that your assessments can rely on.
Bring one process and its documents. We’ll show you the agents, the controls and the record, running on your data.