Zum Hauptinhalt springen
Glossary

AI agent

Updated on 1 min

An AI agent is an AI system that works through a task over several steps on its own: it plans the approach, calls tools such as search, email, calendars or programming interfaces, evaluates intermediate results and decides on the next step. It is usually built on a language model. Under the AI Act, agents fall under the general definition of an AI system, which explicitly covers varying levels of autonomy (Art. 3(1)).

The more autonomy, the further errors reach. An agent with broad permissions can delete data, send messages or place orders, even after a prompt injection has led it astray. Protection comes from minimal permissions, human approval for consequential actions, complete logs and limits on cost and runtime. Approvals lose their value once they become routine, see T03 Excessive automation (automation bias).

No Cookies!

This wiki collects nothing, bakes nothing and leaves nothing behind. There's nothing to consent to. Privacy doesn't get better than this.