What the EU AI Act means for AI agents
The EU AI Act has no separate rules for AI agents. Duties follow your role, provider or deployer, and the agent’s use, on dates running to 2028.
By Sycamore.
Key takeaways
- The EU AI Act has no separate category for AI agents. The Commission’s AI Act Service Desk says its definitions of an AI system and a general-purpose AI model cover them.
- What the Act requires of an agent depends on the company’s role, provider or deployer, and on the agent’s use. A company that builds an agent for its own use can be both.
- An agent is high-risk only through Annex III, which lists uses such as recruitment, or Annex I, which covers regulated products. High-risk duties apply from 2 December 2027 (Annex III) or 2 August 2028 (Annex I).
- Since 2 August 2026, providers must design AI systems intended to interact directly with people so that those people are told they are dealing with an AI system, unless that is obvious.
- The original prohibitions and AI literacy have applied since 2 February 2025, and two Omnibus prohibitions apply from 2 December 2026. General-purpose AI model duties have applied since 2 August 2025 and sit with the model’s provider.
The EU AI Act does not create a separate category for AI agents. The Commission’s AI Act Service Desk says: “Thus, while AI agents are not a separate category of AI under the AI Act, the definitions of an AI system in Article 3(1) AI Act and of a GPAI model in Article 3(63) AI Act are sufficient to cover AI agents.”1 What a company owes depends on whether it is the agent’s provider or deployer, and on what the agent does. Most prohibitions and AI literacy have applied since 2 February 2025 (two newer prohibitions apply from 2 December 2026), transparency duties since 2 August 2026, and high-risk duties apply from 2 December 2027 or 2 August 2028.23 This guide is a summary, not legal advice.
Does the EU AI Act regulate AI agents?
Yes, through the rules for AI systems and general-purpose AI (GPAI) models. The Service Desk says an agent typically contains at least a GPAI model and constitutes an AI system, and that the rules for both therefore apply to agents.1 It points to these rules:1
- the bans on harmful manipulation and on exploiting vulnerabilities, in Article 5(1), points (a) and (b);
- the transparency rules in Article 50, from 2 August 2026, if the agent is intended to interact with people or generate content;
- the high-risk requirements in Chapter III, from 2 December 2027 or 2 August 2028, if the agent classifies as a high-risk AI system.
The Commission calls its view provisional: “Given that developments related to AI agents are recent and fast evolving, the European Commission’s regulatory considerations are only preliminary at this stage.”1
Are you the provider or the deployer of an agent?
You can be either, or both. A provider develops an AI system, or has one developed, and places it on the market or puts it into service under its own name or trademark. A deployer uses an AI system under its authority. Putting into service includes supplying a system for your own use in the EU.4 A company that builds an agent for its own staff can be both its provider and its deployer.
Under Article 25(1), a deployer becomes the provider of a high-risk system if it puts its name on one, substantially modifies one, or changes the intended purpose of a system, including a general-purpose AI system, so that it becomes high-risk.5 Configuring a general-purpose assistant to screen job applicants can be such a change.
When is an AI agent high-risk?
An agent can be high-risk through one of two routes in Article 6. Either it is a safety component of a product, or a product itself, covered by the EU laws in Annex I and subject to third-party conformity assessment, or it is listed in Annex III.6
Annex III covers eight areas: biometrics; critical infrastructure; education; employment and workers’ management; essential private and public services; law enforcement; migration and border control; and justice and democratic processes.7 Examples relevant to agents:7
- recruitment or selection, including filtering job applications and evaluating candidates (point 4(a));
- monitoring and evaluating workers’ performance and behavior, or allocating tasks based on individual behavior or personal traits (point 4(b));
- evaluating people’s creditworthiness, except for detecting financial fraud (point 5(b)).
Under Article 6(3), an Annex III system is not considered high-risk where it does not pose a significant risk of harm, under conditions such as being intended to perform a narrow procedural task. A provider relying on that exception must document its assessment before placing the system on the market or putting it into service. The exception never covers profiling: “Notwithstanding the first subparagraph, an AI system referred to in Annex III shall always be considered to be high-risk where the AI system performs profiling of natural persons.”6
The Commission lists the high-risk obligations as risk assessment and mitigation, high-quality datasets, logging, documentation, information to deployers, human oversight, and robustness, cybersecurity and accuracy.3 For human oversight under Article 14 and deployer duties under Article 26, see how to govern AI agents in production.
What transparency duties apply to agents?
Article 50 applies to certain systems whether or not they are high-risk, from 2 August 2026.23 Its duties include:8
- Providers must design systems intended to interact directly with people so that those people are informed they are interacting with an AI system, unless that is obvious to a reasonably well-informed person in the context.
- Providers of systems that generate synthetic audio, image, video or text must mark the outputs in a machine-readable format so they are detectable as artificially generated or manipulated.
- Deployers must disclose deep fakes, and AI-generated text published to inform the public on matters of public interest unless it has had human review or editorial control and someone holds editorial responsibility.
- The information must be given in a clear and distinguishable manner, at the latest at the first interaction or exposure.
Content generators placed on the market before 2 August 2026 have until 2 December 2026 to meet the marking duty.9
Who carries the duties for the model underneath an agent?
The model’s provider. Article 53 requires providers of GPAI models to keep technical documentation, give documentation to providers of AI systems that integrate the model, put in place a copyright policy, and publish a summary of the training content.10 These duties have applied since 2 August 2025.3 A company that builds an agent on a third-party model keeps its own duties as the agent’s provider or deployer.
What already applies: AI literacy and most prohibited practices
Both have applied since 2 February 2025, except for two prohibitions added by the Omnibus, which apply from 2 December 2026.3211 As amended by the Omnibus, Article 4 now requires providers and deployers to take measures to support the development of AI literacy of their staff and others who operate or use AI systems on their behalf. It adds: “This obligation does not require providers or deployers to guarantee any specific level of AI literacy of any individual.”12 The Commission summarizes the Omnibus change: “Previous AI literacy requirement for companies is simplified, with the Commission and the Member States taking a stronger role in promoting AI literacy”.13
Article 5 bans certain practices outright. For agents, the Service Desk highlights points (a) and (b).1 Point (a) bans AI systems that deploy subliminal techniques beyond a person’s consciousness or purposefully manipulative or deceptive techniques with the objective or the effect of materially distorting the behavior of a person or a group of persons by appreciably impairing their ability to make an informed decision, causing a decision they would not otherwise have taken, in a manner that causes or is reasonably likely to cause significant harm. Point (b) bans AI systems that exploit vulnerabilities due to age, disability or a specific social or economic situation with the objective or the effect of materially distorting behavior in a manner that causes or is reasonably likely to cause significant harm.11
The Omnibus added two more prohibitions, points (ba) and (bb) of Article 5(1).11 Point (ba) covers AI systems that generate or manipulate realistic images, videos, audio or similar material of an identifiable person’s intimate parts, or of an identifiable person engaged in sexually explicit activities, without that person’s freely given, specific, informed, unambiguous and explicit consent for that generation or manipulation. Point (bb) covers AI systems that generate or manipulate child sexual abuse material or performances within the meaning of Article 2, points (c) and (e), of Directive 2011/93/EU, except where a “without right” defense applies under national law. Article 5(1a) narrows both. Placing such a system on the market or putting it into service is banned only in two cases: where that generation or manipulation is its intended purpose; or where its design, training, architecture, capabilities or user-facing functionalities make it a reasonably foreseeable and reproducible outcome without significant technical modification, and the system lacks reasonable and adequate technical safety measures and other safeguards to reliably prevent it, taking into account reasonably foreseeable misuse, and to correct observed or reported misuse. A deployer is caught only when it uses a system for that purpose. Under Article 5(1b), editing material in a way that does not increase the exposure of depicted intimate parts or alter the nature of depicted sexually explicit activities is not manipulation for point (ba).11 Both points apply from 2 December 2026.2
Fines can reach EUR 35 million or 7% of worldwide annual turnover for prohibited practices, and EUR 15 million or 3% for breaches of other listed duties, including Article 50, whichever is higher in each case.14
What applies when?
The Act entered into force on 1 August 2024 and became applicable on 2 August 2026, with some exceptions, including the dates below.3
| Date | What applies |
|---|---|
| 2 February 2025 | AI literacy (Article 4) and prohibited practices (Article 5), except the two Omnibus additions23 |
| 2 August 2025 | GPAI model obligations and governance rules23 |
| 2 August 2026 | The general application date, including transparency (Article 50)23 |
| 2 December 2026 | Prohibitions added by the Omnibus2; Article 50(2) deadline for older systems9 |
| 2 December 2027 | High-risk duties for Annex III systems2 |
| 2 August 2028 | High-risk duties for Annex I systems2 |
The AI Omnibus, which entered into force on 27 July 2026, extended the high-risk dates.3 In the original 2024 text, the high-risk rules fell under the general date of 2 August 2026, except that Article 6(1) and its corresponding obligations were to apply from 2 August 2027.15
Article 111(2) covers systems already running. Apart from the prohibitions, the Act applies to operators of high-risk systems placed on the market or put into service before the high-risk dates only if their designs change significantly from those dates. Providers and deployers of high-risk systems intended for use by public authorities must comply by 2 August 2030 in any case.9
A checklist for agent programs
- Keep an inventory of every agent you build, buy or let employees create. See how to find and inventory every AI agent in your company.
- Record your role for each agent, provider, deployer or both, using Article 3.4
- Classify the use case against Article 5, Annex III and Annex I, and write down any Article 6(3) assessment.6
- Re-classify when the purpose changes, because that can make you a provider under Article 25(1).5
- Add transparency notices at the first interaction, and mark generated content where Article 50 requires it.8
- Collect model documentation from each GPAI model provider under Article 53.10
- Train the people who run agents, as Article 4 requires, and keep a record.12
- Plan high-risk controls early. For agents you classify as high-risk under Annex III, design logging, oversight and monitoring before 2 December 2027.2 As good practice rather than a legal requirement, also give each agent its own identity; see how to give AI agents identities and permissions.
Where Sycamore fits
Sycamore Govern is an AI governance control plane that finds every AI system an organization builds or buys, approves its use, enforces policy at runtime, and preserves the evidence. Its intake captures purpose, data and tool access, models, decisions and human oversight; Govern assigns risk, routes required reviews, and triggers renewed review on material changes. It keeps a tamper-evident record of approvals, policy checks, access, changes and revocations that can be exported as an evidence pack.
Whether an agent is high-risk under the Act, and which role you hold, remains a legal judgment for your compliance team and counsel.
Frequently asked questions
Should we wait for the high-risk dates before doing anything?
No. Several duties already apply regardless of risk tier. They include AI literacy, the original prohibitions and, for agents that interact with people or generate content, the transparency rules. Classifying every agent also takes time, because someone has to find each one, work out its role and use case, and write down the reasoning. Starting with the inventory and the classification means the high-risk work, if any, can be scoped and planned rather than rushed.
Does an agent used only by our own staff fall under the Act?
Internal use does not take an agent out of scope by itself. Putting a system into service for your own use is one of the ways a company becomes its provider. Whether heavier duties apply then depends on the use case. An internal agent that screens job applicants falls in an Annex III area; one that drafts meeting notes does not.
What should we record for each agent?
Record the owner, the intended purpose, your role, the risk tier and the reasoning behind it, the model and its provider, the people who interact with the agent and how they are told, and the date of the next review. Keep the reasoning, not only the conclusion. If a regulator or auditor asks why an agent is not high-risk, the answer should already be written down.
Sources
-
European Commission, AI Act Service Desk, “Frequently asked questions”, entry “How are AI agents addressed within the AI Act?”, https://ai-act-service-desk.ec.europa.eu/en/faq
-
European Commission, AI Act Service Desk, “Article 113: Entry into force and application”, https://ai-act-service-desk.ec.europa.eu/en/ai-act/article-113
-
European Commission, “AI Act”, Shaping Europe’s digital future, https://digital-strategy.ec.europa.eu/en/policies/regulatory-framework-ai
-
European Commission, AI Act Service Desk, “Article 3: Definitions”, points (3), (4) and (11), https://ai-act-service-desk.ec.europa.eu/en/ai-act/article-3
-
European Commission, AI Act Service Desk, “Article 25: Responsibilities along the AI value chain”, https://ai-act-service-desk.ec.europa.eu/en/ai-act/article-25
-
European Commission, AI Act Service Desk, “Article 6: Classification rules for high-risk AI systems”, https://ai-act-service-desk.ec.europa.eu/en/ai-act/article-6
-
European Commission, AI Act Service Desk, “Annex III: High-risk AI systems referred to in Article 6(2)”, https://ai-act-service-desk.ec.europa.eu/en/ai-act/annex-3
-
European Commission, AI Act Service Desk, “Article 50: Transparency obligations for providers and deployers of certain AI systems”, https://ai-act-service-desk.ec.europa.eu/en/ai-act/article-50
-
European Commission, AI Act Service Desk, “Article 111: AI systems already placed on the market or put into service and general-purpose AI models already placed on the marked”, https://ai-act-service-desk.ec.europa.eu/en/ai-act/article-111
-
European Commission, AI Act Service Desk, “Article 53: Obligations for providers of general-purpose AI models”, https://ai-act-service-desk.ec.europa.eu/en/ai-act/article-53
-
European Commission, AI Act Service Desk, “Article 5: Prohibited AI practices”, https://ai-act-service-desk.ec.europa.eu/en/ai-act/article-5
-
European Commission, AI Act Service Desk, “Article 4: AI literacy”, https://ai-act-service-desk.ec.europa.eu/en/ai-act/article-4
-
European Commission, “AI Omnibus enters into force”, https://digital-strategy.ec.europa.eu/en/news/ai-omnibus-enters-force
-
European Commission, AI Act Service Desk, “Article 99: Penalties”, https://ai-act-service-desk.ec.europa.eu/en/ai-act/article-99
-
Future of Life Institute, EU Artificial Intelligence Act, “Article 113: Entry into force and application” (shows the original wording alongside the amended text), https://artificialintelligenceact.eu/article/113/