Control · C-094 · Human Oversight & Contestability
Human override and stop mechanism
A designated person can pause, override or shut down an AI system at any time through a tested mechanism, with the authority to do so settled in advance.
- Control type
- Corrective
- Risks addressed
- 6
- Audience
- Heads-of · Practitioner
- Frameworks
- Principle 4 · Article 14 · MANAGE 4
- Responsible AI domain
- Lifecycle stage
- Reviewed
- 2026-11-11
What it does
It guarantees that a person can stop the system. A mechanism exists to halt the system or a part of it, to override an individual output or decision, and for agents to interrupt work in progress and prevent further actions. The people permitted to use it are named, they have authority to act without seeking approval in an emergency, and they are reachable at all hours for systems that run at all hours. The system owner is accountable; engineering builds the mechanism. It is built at design and available throughout operation.
How to implement it
- Design a stop control that works independently of the system it stops.
- Provide graded options: disable a feature, pause an agent, switch to fallback, shut down.
- Let reviewers override individual outputs, and record each override.
- Name who may invoke the mechanism, and give them standing authority.
- For agents, revoke credentials and halt queued actions as part of the stop.
- Test the mechanism on a schedule, and time how long it takes.
Evidence it operates
Design: documented stop and override mechanisms for each system; named authority. Operation: test records with timings; logs of overrides and stops; an instance of use in an incident; out-of-hours contact arrangements.
Where it fits
Built at design, used in operation and in incidents. It is a primary control for inability to stop or roll back AI, inadequate human oversight design, cascading failures in agent workflows and rogue agent behaviour. It depends on fallback and manual workaround for something to switch to.
Limits
Stopping a system does not undo what it has already done. People hesitate to stop a system the business depends on, and more so when there is no fallback. A stop control inside a supplier's product is only as fast as the supplier. For actions that take effect in milliseconds, a human stop comes after the event.
Risks this mitigates
Primary mappings
- Cascading failures in agent workflows High
- Lets a person halt the whole workflow. An error or compromise in one agent or step propagates through chained agents and automated workflows faster than people can notice or intervene.
- Inability to stop or roll back AI High
- Provides a tested way to pause or shut the system down. A misbehaving system cannot be disabled, reverted to a known good version, or have its actions reversed quickly enough to limit harm.
- Inadequate human oversight design High
- Provides a tested means to pause or override. An AI system is built with no human oversight, or with an oversight step that lacks the time, information, competence or authority to change an outcome.
- Loss of control of advanced AI Medium
- Maintains the ability to halt a system. A highly capable AI system pursues objectives or acquires capabilities beyond the oversight of its developer or deployer, and cannot readily be corrected or stopped.
- Rogue agent behaviour High
- Allows the agent to be halted at once. An agent departs from its intended function or acts deceptively or outside its mandate, through misaligned objectives, emergent behaviour or a compromise that persists, while each action appears legitimate.
Supporting mappings
- Algorithmic collusion and market manipulation High
- Lets a person halt the algorithm at once. Pricing or trading algorithms coordinate with competitors' algorithms or distort a market, without any human agreement to do so.
Framework and clause references
| Framework | Clause | Title |
|---|---|---|
| Code of Practice for the Cyber Security of AI | Principle 4 | Enable human responsibility for AI systems Human responsibility. |
| EU AI Act | Article 14 | Human oversight The ability to interrupt the system. |
| NIST AI Risk Management Framework (AI RMF 1.0) | MANAGE 4 | Risk response, recovery and communication plans are documented and monitored Mechanisms to deactivate. |
Related terms
How to cite
Axiom Verity (2026). Human override and stop mechanism. Reference Library, C-094. https://axiom-verity.com/controls/human-override-and-stop-mechanism. Reviewed 2026-11-11.
Found an error or a better source? Report a correction.