The AI Agent Permissions Review
Template
Review what an AI agent can see, do and change before it gets real permissions, and decide which actions need a human in front of them.
Identity and ownership
- Give the agent its own identity rather than borrowing a user's credentials.
- Name the person accountable for the agent's behaviour.
- Name who approves changes to its tools, instructions and permissions.
Tools and permissions
- List every tool the agent can call and the data each one exposes.
- Separate read access from write access for each tool.
- Remove any permission the agent does not need for its defined task.
Actions needing approval
- Mark actions that are hard to reverse: payments, deletions, external messages, production data changes.
- Decide which of these need human confirmation every time.
- Check that the person confirming sees enough context to make a real decision.
Limits and logging
- Set limits on loops, spend and tool calls per task.
- Treat stored memory and retrieved content as untrusted input.
- Log each tool call with its inputs and outcome so incidents can be reconstructed.
Start a conversation
Have a product or AI decision to make?
Useful first calls usually start with one unclear decision, a deadline and a team that needs a practical next move.
Tell us about it