GOV-06: Per-action policy
In short: Check each agent action against policy at the moment it happens, using the agent, the person behind it, the tool and the data involved. Access granted at login says little about what an agent will do an hour later. Per-action checks catch the risky call itself.
What does GOV-06 require?
Check each agent action against policy at the moment it happens, using the agent, the person behind it, the tool and the data involved.
Why does it matter?
Access granted at login says little about what an agent will do an hour later. Per-action checks catch the risky call itself.
How do you test it?
- Trigger an action the policy should deny and confirm it is blocked.
- Check the decision log shows agent, person, tool and data.
- Change a policy and confirm the next call uses the new rule.
Which platforms document GOV-06?
Documented by NewCore, Silverfort, Astrix Security, Cyera Identity, Descope and Aembit.
- 9.10NewCore documentedSource: NewCore: Secure; NewCore: Govern, read
- 8.35Silverfort documentedSource: Silverfort: AI Agent Security, read
- 7.30Astrix Security documentedSource: Astrix at RSAC 2026, read
- 7.30Cyera Identity documentedSource: Oasis Agentic Access Management, read
- 6.85Descope documentedSource: Descope Agentic Identity Hub, read
- 6.55Aembit documentedSource: Aembit IAM for Agentic AI is generally available, read
- 6.55SailPoint Entro partly documentedSource: SailPoint agentic AI security, read
- 5.50Idira partly documentedSource: CyberArk Secure AI architecture docs, read
- 6.55Token Security not found in public materialChecked: Token Security home, read
- 6.35Clutch Security not found in public materialChecked: Clutch Security platform, read
Which framework references apply?
- NIST NCCoE agent identityProject scope: authorization: How AI agents can be authorized. Source: NIST NCCoE: comments on the agentic AI identity concept paper
- OWASP Agentic Top 10ASI02 Tool Misuse: Agents bending legitimate tools into harmful outputs. Source: OWASP Top 10 for Agentic Applications 2026
Which controls are related?
- GOV-07Human approval for high-risk actions
- GOV-09Policy testing before go-live
- GOV-10Tool allow-listing and hiding
Glossary: agent governance, per-action policy, allow ask deny, mcp gateway, policy enforcement point, conditional access.
Reviewed by The Register desk