Anthropic clarifies rules for autonomous AI use

Anthropic published a revised Usage Policy on October 8, with an effective date of November 12, 2026. Anthropic says many of the revisions clarify existing requirements rather than introducing new restrictions. For organisations operating Claude-based agents, this is a governance development, not an agent feature release or independent security certification.

The company highlights high-risk uses that may influence health, finances, legal rights, livelihoods or essential services. Anthropic requires a qualified human reviewer who can examine and, where necessary, change recommendations. People affected by those decisions should be informed about AI involvement.

The update also addresses models connected to hardware capable of autonomous physical action that might cause injury. Anthropic says a qualified operator must be able to observe and stop the equipment, and the system should remain safe if disconnected. This is a usage-policy condition: it is not evidence that Claude Cowork or Claude Code implements a particular emergency-stop control.

Other revisions consolidate restrictions on deceptive activity and clarify surveillance, weapons-related uses and supported regions. Procurement teams should read the new policy alongside service terms and actual technical deployment documentation. Policy scope does not replace proof of tool permissions, audit logging, retention or processing terms.

AgentenCode links the change to the Claude ecosystem while distinguishing policy publication from its future effective date. No AgentenTrust signal is automatically modified merely because the provider updated its policy.