Anthropic has updated its usage policy to bar users from subjecting Claude to sustained and needless abusive or cruel behaviour. The change was announced on October 8 and takes effect on November 12.

The company says the rule applies only in extreme cases, where users repeatedly act cruelly toward the model with no discernible purpose. It does not cover ordinary frustration, pushback, dark creative themes, or model testing and research.

Claude models are already able to end conversations with persistently abusive users. Anthropic says ending the chat will remain the main enforcement tool, with other conversations on the account unaffected.

The capability was introduced in August 2025 as part of Anthropic’s exploratory research into AI welfare. The company found that Claude Opus 4 showed a consistent aversion to harm and apparent distress in abusive exchanges.

The wider policy update also reorganises related clauses. New sections cover deceptive campaigns, including fake reviews and astroturfing, and a narrower elections section bars deceiving voters.

Anthropic stresses it is not claiming Claude is conscious. It calls the rule a low-cost precaution drawn from its welfare research, now turned into an explicit user-facing restriction.