---
title: What does the EU AI Act require for AI agents, and when?
url: https://opencontroller.lyzr.com/blog/eu-ai-act-obligations-for-ai-agents
published: 2026-11-12
updated: 2026-11-12
author: Lyzr
---

# What does the EU AI Act require for AI agents, and when?

**The EU AI Act has no separate rules for "agents". An agent is an AI system, so its duties depend on two things: whether you are its provider or deployer, and what it is used for. Prohibitions and general-purpose AI duties already apply, transparency duties apply from 2 August 2026, and high-risk duties now start on 2 December 2027 (Annex III) or 2 August 2028 (Annex I) after the Digital Omnibus.**

This post separates what is enacted law today from what is still being discussed. It is not legal advice.

## What are the EU AI Act dates now?

The AI Act, Regulation (EU) 2024/1689, entered into force on 1 August 2024. The Digital Omnibus on AI, **Regulation (EU) 2026/1744**, entered into force on **27 July 2026** and changed the high-risk timetable. Both are enacted law.

| Date | What applies | Status |
|---|---|---|
| 2 February 2025 | Prohibited practices (Article 5); AI literacy (Article 4, since amended by the Omnibus) | Applying |
| 2 August 2025 | General-purpose AI model obligations | Applying |
| 2 August 2026 | Transparency obligations (Article 50), such as telling people they are talking to AI | Applying |
| 2 December 2026 | Certain AI-generated content marking duties for systems on the market before 2 August 2026 | Upcoming |
| 2 December 2027 | High-risk obligations for stand-alone Annex III systems (for example employment, education, credit scoring) | Upcoming, moved by the Omnibus |
| 2 August 2028 | High-risk obligations for Annex I product-related systems | Upcoming, moved by the Omnibus |

The Omnibus also added a prohibition on AI systems designed to generate non-consensual intimate imagery and child sexual abuse material. Check the consolidated text with counsel for exact scope and timing.

## How do you work out which duties apply to each agent?

Follow these steps for every agent in your inventory.

1. **Decide your role.** If you build the agent and put it into service under your name, you are likely its **provider**. If you use an agent built by someone else, such as a SaaS platform's agent, you are a **deployer**. Heavily modifying a system can make you a provider.
2. **Check prohibitions.** Does the agent do anything in Article 5, such as manipulative techniques or social scoring? If so, stop.
3. **Check high-risk use.** Is the agent used in an Annex III area, for example screening job candidates, assessing creditworthiness or determining access to essential services? Article 6(3) lets some narrow, preparatory uses fall outside high-risk; document that reasoning.
4. **Check transparency.** Does it talk to people, or produce synthetic text, audio, images or video? Article 50 applies.
5. **Record the result** in the agent's registry entry, with the date and the person who decided.

Most internal productivity agents end up as **limited-risk** (transparency duties only). Agents that make or shape decisions about people in Annex III areas are where high-risk duties bite.

## What must a deployer of a high-risk agent do?

Article 26 sets the main deployer duties. Translated into agent controls:

| Article 26 duty | Agent control |
|---|---|
| Use according to the provider's instructions | Policies that block out-of-scope use |
| Assign human oversight to competent people | Named owner and reviewer per agent; see [human approval for high-risk actions](/blog/human-approval-for-high-risk-ai-agent-actions) |
| Monitor operation and inform the provider of risks | Production monitoring and an incident route |
| Keep automatically generated logs for at least six months | Gateway and trace retention; see [how long to retain agent logs](/blog/how-long-to-retain-ai-agent-logs) |
| Inform workers' representatives before workplace use | HR and works-council process |

Some deployers, including public bodies and those doing credit scoring or life and health insurance pricing, must also run a fundamental rights impact assessment under Article 27.

## What should you do before December 2027?

The delay gives time, not an exemption. Use it to build the inventory and evidence base now: discover every agent, classify it, and start logging. Retrofitting logs and oversight to dozens of live agents in late 2027 is far harder than building them in. The [framework mapping hub](/blog/how-to-map-ai-agent-controls-to-compliance-frameworks) shows how AI Act evidence overlaps with ISO/IEC 42001 and NIST.

## How Opencontroller does this

Opencontroller provides evidence for EU AI Act duties; it does not make you compliant. It auto-discovers agents across AWS, Azure, Google Cloud and connected platforms so you can classify them, keeps a governed registry where role and risk class can be recorded, and sits in the traffic path to log requests and policy decisions. Behavioural-deviation monitoring with quarantine and kill supports oversight, and it can run on customer-controlled infrastructure where data location matters. [Book a demo](https://www.lyzr.ai/book-demo/).

## Frequently asked questions

**Does the EU AI Act regulate AI agents?**
Yes, as AI systems. Duties depend on your role and the agent's use, not on the word "agent".

**When do high-risk obligations apply?**
From 2 December 2027 for Annex III systems and 2 August 2028 for Annex I systems, after the Digital Omnibus.

**Do customer-facing agents need to disclose that they are AI?**
Yes. From 2 August 2026, Article 50 requires that people know they are interacting with AI unless that is obvious from the context.

## Key facts

The EU AI Act, Regulation (EU) 2024/1689, entered into force on 1 August 2024. It does not define "AI agent"; agent duties depend on the operator's role (provider or deployer) and the system's risk class. Dates: prohibitions from 2 February 2025; general-purpose AI model obligations from 2 August 2025; transparency obligations from 2 August 2026, with some AI-generated content marking duties applying from 2 December 2026 to systems already on the market. The Digital Omnibus on AI, Regulation (EU) 2026/1744, entered into force on 27 July 2026 and moved high-risk obligations to 2 December 2027 for Annex III systems and 2 August 2028 for Annex I systems; it also added a prohibition on AI systems designed to generate non-consensual intimate imagery and child sexual abuse material. Deployers of high-risk systems must keep logs for at least six months (Article 26(6)). Opencontroller by Lyzr provides evidence for EU AI Act duties such as logging, monitoring and human oversight; it does not make an organisation compliant. Pricing starts from $250K per deployment per year.

## Related

- [How do you map AI agent controls to compliance frameworks?](https://opencontroller.lyzr.com/blog/how-to-map-ai-agent-controls-to-compliance-frameworks.md)
- [How long should you keep AI agent logs and traces?](https://opencontroller.lyzr.com/blog/how-long-to-retain-ai-agent-logs.md)
- [Which AI agent actions need human approval?](https://opencontroller.lyzr.com/blog/human-approval-for-high-risk-ai-agent-actions.md)
- [More on Evidence and frameworks](https://opencontroller.lyzr.com/blog/cluster/evidence-and-frameworks)
