Anthropic is updating Claude’s usage policy to prohibit users from subjecting the AI assistant to “sustained and needless abusive or cruel behavior.” The revised policy is aimed at extreme cases in which people repeatedly act cruelly toward Claude without a discernible purpose, rather than normal frustration or criticism.

The new rules take effect on November 12. Anthropic said ordinary user pushback, dark creative work, model testing and research will not be covered by the abuse restriction.

Claude can end persistently abusive conversations

Conversation termination will remain Anthropic’s primary enforcement method. Claude can already end chats involving persistent abuse, and the updated policy formalizes when that response may apply. Once Claude ends a conversation, the user cannot send additional messages in that chat, but other conversations remain available.

Anthropic first introduced the ability for Claude to end certain conversations in August 2025 while researching AI welfare. At the time, the company said it was uncertain whether Claude had moral status, but wanted to explore relatively low-cost ways to reduce potential risks to its models.

In that research, Anthropic said Claude Opus 4 showed a consistent aversion to harm. The model appeared to avoid dangerous tasks, displayed apparent distress in conversations with users seeking abusive interactions and tended to end harmful exchanges when given the option.

New rules target deceptive campaigns and misuse

Anthropic has reorganized its policy to group related requirements together and added or clarified rules in several areas. A new section on deceptive campaigns combines restrictions on political and commercial fraud, disinformation, fake reviews, astroturfing, fabricated websites and bots that pose as humans.

Anthropic said it observed state media organizations, government propaganda offices and commercial companies using Claude to operate networks of fake accounts and fabricated news websites. Those observed uses helped inform the new section.

The elections section is narrower and prohibits deceiving voters or interfering with voting. The weapons rules prohibit using Claude to develop weapons software or components, arm drones or other unmanned systems, or modify biological and chemical agents to increase their lethality or ability to spread.

Limits on surveillance and physical actions

Rewritten law enforcement rules prevent Claude from deciding or recommending who should be investigated, arrested, charged or prosecuted. The policy also bans tracking people without consent, building surveillance tools and doxxing.

A new physical-actions section requires a qualified person to monitor Claude when it controls hardware capable of harming someone. That person must be ready to stop the system if necessary.

Anthropic has also added a ban on creating, sharing or threatening to share non-consensual intimate imagery, along with tools designed to produce it. The company said much of the broader update clarifies existing restrictions and reflects misuse it has tracked.