Anthropic, the U.S. developer of the Claude AI models, released a new edition of its usage policy on Oct. 8. The document takes effect on Nov. 12 and for the first time includes a ban on “sustained and needless abusive or cruel behavior toward our models.”
The company said the rule is intended only for extreme cases in which a user behaves cruelly again and again without any apparent purpose. Ordinary dissatisfaction with answers, objections, dark themes in creative tasks, and testing and research on the models are not considered violations.
Claude’s ability to end a conversation on its own will remain the main way the rule is enforced. Anthropic introduced that feature in Claude.ai and Claude Code in August 2025 as part of research into so-called “model welfare,” stressing at the time that the moral status of language models remains deeply uncertain.
Surveillance and policing
The new version also spells out restrictions on surveillance more precisely. Tracking people without their consent is prohibited, whether in real time or by analyzing previously collected data. Claude may not be used to decide, or to recommend, whom law enforcement should vet, detain or charge, nor to build surveillance tools. Uses such as content moderation, journalism and legal research remain permitted.
Unsupported regions
Anthropic also clarified how it applies restrictions for unsupported regions: the service may not be used by people physically located in those regions, or by companies registered there or majority-owned by individuals or entities from those regions. Russia is not among the countries where Claude is available.
