The EU AI Act and AI agents.
The Act regulates AI systems by what they are used for, not by the model behind them. For a team running agents, four of its requirements land directly on how those agents are given access.
- The Act classifies by use, not by model. The same agent is low or high risk depending on what you point it at.
- High-risk systems must keep automatically generated logs, traceable to events. For an agent that is an identity question first.
- Human oversight must be able to intervene, which in practice means being able to revoke what the agent runs as.
- Deployers carry duties too. Buying the agent does not move the obligation to the vendor.
Published Last reviewed
What it actually requires of agent access.
Narrowed to the part that touches what an agent can reach. The full text covers much more.
The Act regulates AI systems by use, not by how impressive the model is. An agent that triages support tickets and an agent that screens job applicants sit in different tiers for the same technology, and the second one is high risk under Annex III.
High-risk systems must keep automatically generated logs over their lifetime, traceable to the events that produced them. For an agent this means the record has to say which identity made which call, which is an identity design question before it is a logging one.
Oversight that cannot intervene is not oversight. In practice the ability to stop an agent mid-task is the ability to revoke what it runs as, so the control lives with the credential rather than with the interface.
Buying an agent does not move the duty to the vendor. A deployer has to use the system as intended, keep the logs it generates and assign oversight to people with the competence and the authority to act.
Questions, answered.
Short answers on eu ai act and ai agents.
Does the EU AI Act apply to an internal AI agent?
It can. The Act applies to AI systems placed on the market or put into service in the EU, including systems an organisation builds for its own use. What decides the obligations is the use case: an internal agent doing Annex III work, such as screening applicants, carries high-risk duties even though nobody sells it.
What does the Act say about agent permissions?
Nothing directly, and that is the point worth noticing. It asks for record keeping, human oversight and risk management, and each of those becomes an access control question once the system acts on its own. Logs that cannot say which identity made a call do not satisfy traceability.
Who is responsible, the vendor or us?
Both, in different roles. The provider carries the conformity obligations for the system. The deployer has to use it as intended, keep the logs it generates, and assign oversight to people who can actually intervene.
Keep reading.
The pages that belong to the same picture.
See what your agents can reach. Before someone else does.
Book a demo and we walk through the access paths in an environment like yours.