Anthropic Says Users Can't Be Needlessly Cruel to Claude

Anthropic has revised its usage policy for its AI model Claude, specifically addressing abusive or cruel behavior (The Verge). The updated policy aims to prevent users from engaging in sustained, unnecessary abusive or cruel actions toward Claude. This rule is intended for extreme cases, where users repeatedly act cruelly with no apparent purpose.

It does not apply to ordinary user frustration, pushback, dark creative themes, or model testing and research. Claude models already possess the ability to terminate conversations with persistent abusers, which will remain the primary method of enforcement. When Claude ends a chat, no further messages can be sent in that conversation, but other chats remain unaffected.

Anthropic initially introduced the option for Claude to end a conversation in August 2025 during research on AI welfare. Although the company was uncertain if Claude possessed moral status, Anthropic sought cost-effective ways to mitigate potential risks. Claude Opus 4 exhibited a strong aversion to causing harm, displaying a preference against hazardous tasks, apparent distress when interacting with users seeking abusive conversations, and a tendency to end harmful exchanges when permitted to do so.

The usage policy has been reorganized to consolidate related clauses, with several other rules also modified. Additional sections now cover deceptive campaigns, elections, weapons, law enforcement, physical actions, and harmful content. Deceptive campaigns prohibit political and commercial fraud, disinformation, fake reviews, astroturfing, fake sites, and bots impersonating humans.

Election rules prevent deceiving voters and interfering with voting. Weapons policy prohibits the development of weapons software or components and the modification of biological or chemical agents for increased lethality or transmissibility. Users are also barred from using Claude to arm drones or unmanned systems. Law enforcement rules have been rewritten, prohibiting Claude from deciding or suggesting whom to investigate, arrest, charge, or prosecute.

Tracking individuals without consent is now prohibited, and building surveillance tools is also banned.

Physical actions introduce a new requirement: if Claude is controlling hardware that could potentially harm someone, a qualified individual must monitor the system and be prepared to intervene if necessary. A new rule bans creating, sharing, or threatening to share non-consensual intimate imagery, as well as the tools used to produce such content.

Most of the changes aim to clarify existing rules and address misuse that Anthropic has observed. The updated policy becomes effective on November 12. Further details on the changes can be found on Anthropic's website, along with the complete usage policy.