Trust Center
Human-in-the-loop controls
Human authority is preserved at the policy layer. High-risk actions require explicit approval, and all overrides are captured in audit records.
Control model
- Risk-based action tiers route low-risk tasks to automation and hold high-risk tasks for human approval.
- Approval records include reviewer identity, timestamp, rationale, and linked evidence IDs.
- Escalation chains define on-call operational owner, governance owner, and executive contact.
- Emergency stop can revoke all active agent tokens and block new autonomous runs.