AI agents and automated actions

Overview

An agent can act, not just answer: it can send messages, change records or spend money. The main control is limiting what it can reach and requiring approval for actions that are hard to undo.

Key risks

• Actions beyond intended scope
• Runaway cost or loops
• Actions that cannot be undone
• Prompt injection that triggers actions

What to do now

• Give the least permissions needed.
• Put a person in the loop for high-impact actions.
• Log every action and test a stop mechanism.
• Set rate and spend limits.

Questions to ask

• What actions can the agent take, and which need human approval?
• What is the smallest set of permissions it needs?
• Can we see and replay what it did?
• How do we stop it quickly if it misbehaves?
• How do we test for hidden instructions in content it reads?

Related tools

• Agent checklist

Region rules: EU · US · UK · India · Global standards.

Reviewed 3 Oct 2026. Sources: European Commission: AI Act overview.