Audience
Product, engineering, security, privacy, risk and operations leaders
AI Agents & Governance · Updated August 8, 2026
By PlanckCyber
AI Agent Control Matrix
Tools, permissions, human authority and evidence for agentic systems.
How to use this resource
Create one row for every tool or material action. “The agent needs access” is not a control decision.
Important: Does not replace legal, privacy, security or regulatory review.
When to use it
An agent will access tools, data or actions.
Objective
Make authority and control decisions explicit before implementation.
System identity
- Agent / system name
- Workflow
- Operating owner
- Technical owner
- Risk owner
- Environment
Authority matrix
- For each tool or action, record its purpose, data, permission, human approval, audit record and revoke / rollback method.
Control questions
- Least privilege is documented.
- External content is treated as untrusted.
- Prompt / instruction hierarchy is protected.
- Outputs are validated before material use.
- High-risk actions require human authority.
- Rate, cost and action limits exist.
- Secrets are not exposed to the model unnecessarily.
- Logging excludes prohibited sensitive data.
- Incident response and shutdown are tested.
- Model / tool changes trigger regression testing.
Approval
- Open risks
- Conditions before release
- Approval owner
- Date
- Next review
Start with the problem
Have a problem AI might solve?
You do not need a specification. Tell us what you are trying to improve.