Being mean to AI could get you banned as Anthropic introduces new rules against Claude abuse

WTF?! You may have heard about being kind to chatbots in case they remember you once the machines take over. It appears that Anthropic is heeding this warning, banning users from being excessively abusive or cruel to its AI models.

Anthropic’s first changes to its usage policy in over a year include new restrictions on abusive behavior. The updated policy prohibits “sustained and unnecessary abusive or cruel behavior” toward its models.

Don’t worry if you’re a Claude user who sometimes gets angry with the chatbot and throws a few curse words at it. The policy update only applies in extreme cases, in which users act repeatedly and cruelly towards models for no discernible reason.

Anthropic also writes that the policy does not apply to common expressions of user frustration, opt-outs, dark creative themes, or model testing and research.

This change is an expansion of a feature introduced in August 2025. Anthropic said at the time that Claude would be allowed to end conversations with users in rare and extreme cases of consistently harmful or abusive interactions.

The company said this feature was developed primarily as part of its exploratory work on the potential well-being of AI, although it has broader relevance to model alignment and safeguards.

Anthropic says Claude’s ability to stop abusive interactions will remain the primary enforcement mechanism. Its broader policy allows for warnings, restrictions, suspensions, or termination of access, so bans are a possibility.

In addition to its new rules against malice toward Claude, Anthropic is clarifying rules regarding fake accounts and propaganda campaigns, weapons software, and surveillance.

The section on deceptive activities brings together existing restrictions on political and commercial manipulation, including networks of fake accounts and fabricated news sites. Claude also cannot recommend who the police should investigate or arrest.

The company is also dropping its blanket ban on personalized voter and campaign targeting, allowing legitimate civic work while maintaining restrictions on deception and voter suppression.

There are clearer requirements for human oversight in areas such as health and finance, as well as new safeguards for Claude to control potentially dangerous material. A qualified operator must be able to monitor and shut down the equipment, which must remain safe if Claude disconnects.

Other changes address non-consensual intimate images and clarify access restrictions involving unsupported regions, including companies majority owned or controlled by individuals or entities based in those regions.

The updated policy goes into effect on November 12, giving users some time to work on their chatbot ways.