Suspension, revocation & reinstatement
Governance without an enforcement pathway is documentation. The framework's teeth are three rules: scope breaches suspend automatically, revocation is a human decision on the record, and the way back is through proctoring — never through a quiet re-enable.
Rule one: breach suspends, automatically
An out-of-scope action suspends the agent the moment it is logged — no meeting, no judgment call, no grace period. This is the agent analogue of a model breaching its equity threshold: the automatic response is what makes the scope real rather than aspirational. False positives are cheap (a wrongly suspended agent loses hours); false negatives are the incident report.
Rule two: revocation is decided, suspension is triggered
Suspension is mechanical; revocation is judgment. A suspended agent gets a human review: was the breach a scope-definition error (fix the scope, reinstate), a vendor change that altered behavior (a contract conversation), or conduct that ends the deployment (revoke)? The decision, its maker, and its rationale go on the permanent record — the same record a surveyor, a court, or your own future committee will read.
Rule three: reinstatement means proctoring again
A reinstated agent returns to proctoring, not to its old privileges. Whatever caused the suspension changed the trust calculus, and the proctoring period is where trust is rebuilt on evidence. Organizations that skip this step have converted their enforcement pathway into a light switch — and the record will show exactly that.
All five components run natively in HAI-OS; the public demo shows the machinery live.