Anthropic revises Claude rules on deception, elections and physical agents
The policy clarifies limits on influence operations, weapons software and surveillance, while adding conditions for physical agents and extreme abuse toward Claude.
Rules for more autonomous use
Anthropic published a revised Usage Policy for Claude on October 8, with an effective date of November 12. The company says many changes clarify restrictions it already enforced. A new consolidated section covers deceptive campaigns, including fake accounts and fabricated news sites. The election section now focuses on deceiving voters or disrupting voting; Anthropic also says it removed a blanket ban on personalized campaign targeting because that rule caught legitimate civic work. Deceptive targeting and misuse of personal data remain prohibited under other sections.
Physical controls and an unusual abuse rule
The update makes the company’s weapons-software and surveillance restrictions more explicit. For Claude connected to hardware capable of causing injury, Anthropic now requires an operator who can observe and stop the equipment and a safe state if the model disconnects. It also prohibits sustained, needless abusive behavior toward its models in extreme cases. Anthropic says ordinary frustration, criticism, dark creative themes and research testing are unaffected, and that allowing Claude to end rare abusive conversations remains the main enforcement mechanism. These are policy statements, not evidence that every prohibited use can be detected or prevented.