AIG-023 AI System Override and Safe-State Mechanisms
Description
Each production AI system has a documented and tested mechanism allowing authorised operators to: (a) override or reject individual AI outputs without consequence, (b) temporarily suspend AI-assisted processing and fall back to manual procedures, and (c) fully deactivate the system and bring it to a defined safe state. Override and deactivation procedures are documented, accessible to operators, tested at least annually, and exercisable without requiring vendor involvement for Tier 3 systems.
Rationale
The ability to override and deactivate is a core safety control; AI systems that cannot be safely stopped or overridden are ungovernable.
Framework Mappings (3)
| EU-AI-Art.14.2 | Human Oversight — Capabilities Assigned to Oversight Persons | full |
| EU-AI-Art.15.2 | Accuracy, Robustness and Cybersecurity — Resilience and Fail-Safe Design | partial |
| MANAGE 2.4 | AI System Deactivation and Override Mechanisms | full |
Evidence (1)
Documented and tested override, suspension, and safe-state deactivation procedures for each production AI system, including test results confirming the mechanisms function as designed.
Example: AI System Override Test Record — Content Moderation Engine v3 (Confluence): override test performed 2026-01-15 by Ops Lead, individual output rejection confirmed functional, fallback to manual queue confirmed operational within 2 minutes, full deactivation test result: endpoint offline in 45 seconds, rollback verified
Test: Request override and safe-state test records for a sample of Tier 2+ AI systems. Verify: (1) all three mechanisms are tested (individual output override, suspension/fallback, full deactivation), (2) test was conducted within the last 12 months, (3) test results confirm mechanisms function as designed, (4) for Tier 3 systems, deactivation does not require vendor involvement (confirm from test notes), (5) procedures are documented and accessible to operators.
Questions (2)
Does each production AI system have a tested mechanism allowing authorised operators to override outputs, suspend processing, and fully deactivate the system?
AI systems that cannot be safely stopped or overridden are ungovernable. Override, suspension, and deactivation procedures must be documented, accessible to operators, and tested at least annually — not left to vendor support.
Which of the following override and safe-state capabilities have been tested in the last 12 months for your production AI systems?
All four capabilities should be tested annually. For Tier 3 systems, vendor-independent deactivation is a hard requirement — systems where stopping requires raising a vendor support ticket are not operationally safe.