Anthropic: Users no longer allowed to insult Claude

Anthropic has updated Claude’s usage rules to ban “offensive or cruel behavior” toward the chatbot. The company says the rule is aimed at extreme cases, not ordinary friction. If someone has a good reason to be sharp, or is writing fiction or doing research, that alone won’t get a chat cut off.

Claude has reportedly been able to end conversations since August when users turn hostile for no reason. Anthropic frames the new policy as part of “model welfare,” a term that treats AI systems as something that deserves care and consideration, even as the company says it does not know whether its models are conscious.

That puts Anthropic in a awkwardly philosophical corner of the industry. Microsoft, by contrast, is said to reject the whole idea of model welfare and sees no reason to talk about AI “well-being.” Same technology, very different instincts about what it owes users, and what users owe it back.

The new guidelines also target requests around health and finance, electoral manipulation, and weapons development. Anthropic says the aim is to reduce misuse. It also says it has repeatedly found Claude being used for surveillance and for identifying and tracking political opponents, which is now explicitly banned.

There are still carve-outs. Government institutions, journalists, and researchers can keep using Claude in those areas within a defined framework. And Anthropic says it can override all of these rules in contracts with certain government clients if it thinks the safeguards are strong enough to limit harm.