Anthropic announced an updated Usage Policy on October 8 that prohibits sustained, needless abusive or cruel behavior toward its models. The rule takes effect November 12 and is intended for extreme cases involving repeated cruelty with no discernible purpose. Ordinary frustration, pushback, dark creative themes, and model testing or research are excluded. Anthropic says ending these conversations will remain the main enforcement mechanism.
The change builds on a feature introduced in August 2025 that allows Claude to end rare, persistently abusive conversations. Anthropic described that intervention as a low-cost precaution within its work on possible model welfare, while acknowledging substantial uncertainty about whether models have morally relevant experiences.
For welfare preparedness, this offers a practical example of setting boundaries around how people treat machine intelligence before questions about consciousness have been settled. We can examine whether a precaution is reasonable, proportionate, and useful while those questions remain open.
The policy does not establish that Claude is conscious or can suffer; it does not establish the opposite either. Nor does adopting a precaution mean the evidence is evenly balanced. Whether this particular intervention protects welfare, and how it works in practice, remain questions for investigation.
Sources: https://www.anthropic.com/news/2026-usage-policy-update and https://www.anthropic.com/research/end-subset-conversations .