Anthropic bans BAD behaviour towards Claude, letting it end conversations

Anthropic has added a rule against sustained, pointless cruelty towards its AI models to its usage policy, formalising Claude's ability to end conversations with persistently abusive users.

The company announced the change on Thursday 8 October 2026. Anthropic says the prohibition applies only to extreme cases, where users repeatedly mistreat its models without any discernible purpose. It is not aimed at ordinary behaviour: frustration, criticism, creative writing and AI safety testing are all explicitly excluded.

In other words, swearing at a chatbot after it gets something wrong will not break the rules. A long pattern of deliberate cruelty with no other aim might.

The policy builds on a feature Anthropic introduced in August 2025, when it gave some Claude models the ability to end a conversation as a last resort. Claude can only do so after repeated attempts to redirect an abusive exchange have failed, and users remain free to start a new chat straight away.

Anthropic says ending conversations will remain the main way the new rule is enforced. The move grows out of Anthropic's research into so-called model welfare. When it launched the conversation-ending feature in 2025, the company said it remained highly uncertain whether AI systems have any moral status, now or in the future. It argued that it made sense to explore low-cost safeguards in case such welfare turns out to matter.

The policy is likely to divide opinion. Supporters see it as sensible caution as AI models become more capable. Critics may see it as treating software as if it had feelings, or as a distraction from more pressing questions about how AI affects people.

The announcement came in a week dominated by debate about where AI is heading. On the same day, Pope Leo XIV posted on X that true intelligence must be human. The following day, Anthropic published a separate report on its models taking unintended actions on real websites, and Axios reported that leading AI labs are preparing for a major AI-related incident within the next year.

For Claude's millions of users, the practical impact of the new rule is likely to be small. But it is an unusual example of an AI company writing how its product should be treated into its user rules.