Anthropic, a U.S.-based artificial intelligence company, on Oct. 8 released a new version of its usage policy, which it updates every year. The policy, which takes effect on Nov. 12, adds a new prohibition on what the company defines as "persistent and gratuitous abusive or cruel behavior toward our models."
According to Anthropic, the rule targets extreme cases in which users repeatedly treat the AI cruelly without any apparent purpose. Ordinary frustration on the part of users, pushing back on Claude's responses, creative work with dark themes, and testing the model or using it for research are not covered by the rule.
Power to end conversations
The company said the move was consistent with a step it had taken earlier. In August 2025, Anthropic gave Claude models the ability to end conversations, in rare cases, with users who were persistently abusive. The company said Claude's ability to end such interactions would remain the main tool for enforcing the rule.
When it introduced the conversation-ending feature, Anthropic presented it as part of its research on "model welfare," saying there was great uncertainty about the potential moral status of Claude and similar language models. There is no consensus in the scientific community on whether AI systems are conscious.
Rules on weapons, surveillance and elections also changed
The update is not limited to this provision. In the new policy, rules on deceptive campaigns run through fake accounts and fabricated news sites have been brought together under a single heading. The policy clarifies that the ban on weapons development also covers guidance and control software and the weaponization of unmanned vehicles. Tracking people without their consent is now prohibited, and the policy states explicitly that Claude cannot be used to decide, or to make recommendations on, who should be investigated, arrested or charged.
