Anthropic bans persistent abuse and cruelty toward Claude

What's new

Anthropic, the company behind the Claude chatbot, has updated its usage policy for the first time in about a year. The rules take effect on November 12, 2026. The new line that has drawn the most attention prohibits "sustained and needless abusive or cruel behavior" toward Claude. In other words, persistent and unnecessary abusive or cruel conduct.

Headlines in several outlets say you can be kicked out for insulting Claude. That is a sharp oversimplification. Anthropic itself says the rule applies only in extreme cases, where users repeatedly are cruel to the models with no visible purpose. Ordinary frustration is not covered. The rules apply to all users of Claude.ai and Claude Code, to developers and companies that use Claude through the API, and to customers of third-party products that have Claude built in.

Other parts of the policy have also been tightened or adjusted. The weapons rules now also cover software and components for autonomous weapons. Surveillance is prohibited if it happens without consent, and Claude may not be used to identify who should be investigated or arrested. The election rules have been loosened, so that, for example, associations and election authorities can use targeted messaging for lawful purposes.

What's smart

What's interesting is how the rule ties in with something Anthropic has already introduced. Since the summer of 2025, some Claude models (Opus 4 and 4.1) have been able to end a conversation if the user is persistently harmful or abusive, after Claude has first tried to redirect or calm the situation. Anthropic says this ability remains the primary way of enforcing the rule.

When a conversation is ended, you can't write any more messages in that thread. You keep your account, can start a new conversation, and can edit earlier messages to try a different approach. Anthropic bases this on research into AI welfare. The company remains highly uncertain whether Claude has any kind of moral status, but sees this as a low-cost measure in case it matters.

Cheaper or better?

This isn't a new model, so there is no price, speed or benchmark to compare. No prices are changing. The sources I found don't say whether competitors such as OpenAI and Google have similar rules about behavior toward their chatbots.

It is also unclear when a warning turns into an account suspension. Media outlets report that the penalties range from ended conversations to suspension and permanent bans, but according to the sources, Anthropic has not specified whether an account can actually be blocked solely for this kind of behavior. Automatic blocking has not been introduced for every single insult either.

What it's good for

For most people in Denmark, nothing changes in daily use. If you swear when Claude makes a mistake, you are not affected, according to Anthropic. The same goes for dark themes in creative writing and for testing the model, for example when you try to find its limits.

The rules are most relevant for companies building solutions on top of Claude. For them, it is worth reading the updated rules on weapons, surveillance and autonomous systems before November 12. Among other things, the rules require that qualified operators be able to monitor and stop autonomous systems that could cause harm.