Why AI agent governance needs independent outcome evidence

An AI agent's account of what it intended to do is useful, but it is not independent evidence.

A credible evidence model separates intent, action and independently observed outcome.

ADA's Forge platform provides implemented governance foundations; runtime interception and independent security evaluation remain planned R&D work.

The minimum evidence chain

• Agent and model identity
• Policy and authority
• Attempted action
• Enforcement decision
• Independently observed outcome
• Human or automated decision

What a demonstrator must prove

• Bind identity and policy
• Mediate a constrained tool
• Observe resulting state independently
• Revoke authority
• Emit a verifiable evidence bundle