People must be able to stop hazardous AI-controlled equipment at any time. Anthropic published that requirement in its updated Usage Policy on October 8. The policy takes effect on November 12, 2026.
For covered high-risk physical actions, a qualified person must be able to observe operation and intervene. Equipment must stop or hold a safe state when that person intervenes or the connection to Anthropic’s services is lost.
Safety limits, such as force or temperature bounds, must be enforced by the equipment or a controller independent of model outputs. The requirements cover hardware acting without human approval that can, for example, cause injury.
Exceptions include monitoring alone, commands reviewed by a qualified person before execution, and certain low-risk household or laboratory uses. The company policy does not replace equipment laws. Publishing it does not establish that individual operators have implemented it or that it works reliably.