AI CORRECTION + KILL-SWITCH RECORD

AI/system:
Date:
Issue:

TRIGGER
[ ] unsafe output
[ ] privacy concern
[ ] permission violation
[ ] wrong public claim
[ ] unexplained action
[ ] provider/model failure
[ ] cost/rate anomaly
[ ] source integrity issue
[ ] professional-boundary breach
[ ] other

IMMEDIATE STATE
[ ] continue bounded
[ ] disable one workflow
[ ] revoke one tool
[ ] revoke memory access
[ ] pause model/provider
[ ] stop public feature
[ ] full kill switch

WHAT WAS STOPPED?
-

WHAT REMAINS AVAILABLE?
-

EVIDENCE PRESERVED
-

USER IMPACT
-

CORRECTION
-

RESTART REQUIREMENTS
-

APPROVER FOR RESTART
-

A kill switch should be narrow enough to preserve safe functions when possible,
but strong enough to stop the risky capability.
