Anthropic’s November 12 policy puts hardware safety outside the model
Anthropic’s revised Usage Policy carries a November 12 effective date, bringing a practical checkpoint for developers connecting Claude to equipment. This is the effective date of a policy announced October 8, not a new model release or a new statute. Neither checked page specifies an exact cutover hour. Announcement
For covered high-risk physical actions, the policy requires a qualified person who can observe and stop operation. Equipment must stop or hold a safe state when that person intervenes or the service connection is lost. Safety limits must be enforced by equipment or a controller independent of model output. The scope concerns hardware acting without human approval that can cause specified physical risks, not every ordinary chatbot interaction. Current policy
Why it matters: analysis
The important architectural boundary is between asking an AI to behave safely and making unsafe behavior impossible at the equipment layer. A prompt can help direct a task, but it is not an independent stop mechanism.
For teams testing physical agents, a useful review would record who owns the stop control, what happens after a network failure, and which limits remain enforced if the model emits an unexpected action. Those checks should be demonstrated rather than inferred from a successful demo. They do not replace applicable machinery or workplace rules.
Sources and citations
Revision history
Published .