Anthropic’s new ban on abusive behavior toward Claude

Anthropic’s updated usage policy prohibits “sustained and needless abusive or cruel behavior” toward Claude. The change is one of the policy’s notable additions, alongside new rules addressing high-risk uses of AI. Anthropic said the update reflects new and high-risk cases of misuse, but the restriction on abusive treatment is specifically about how people interact with Claude.

The company had previously addressed harmful interactions through its research into “model welfare.” Last August, Anthropic announced that Claude could end conversations with users who were “persistently harmful or abusive.” That announcement established conversation termination as a way for the model to respond to such users; the updated policy now explicitly prohibits sustained and needless cruelty or abuse toward Claude.

The wording focuses on behavior that is both sustained and needless, rather than describing every unpleasant exchange as a violation. Anthropic’s earlier announcement also referred to persistent harmful or abusive conduct. Together, the policy update and the prior announcement set out the company’s position: persistently abusive interactions are not permitted, and Claude may end a conversation when a user’s behavior is persistently harmful or abusive.

How Anthropic will enforce the rule

Anthropic says terminating conversations remains the primary way it will enforce the restriction on sustained and needless abusive or cruel behavior toward Claude. The company announced last August that Claude could end conversations with users who were persistently harmful or abusive, as part of its research into “model welfare.” The updated policy keeps that measure in place.

That means the central response described in the policy is for Claude to stop engaging when a user’s behavior is persistently harmful or abusive. The rule is not limited to a single unpleasant exchange: it addresses sustained conduct, and the stated enforcement mechanism is conversation termination.

Anthropic’s update comes alongside new restrictions covering high-risk misuse, including election interference, weapons development, surveillance, and health and financial uses. But for abusive treatment of Claude, the policy’s primary enforcement mechanism remains ending the conversation. The company says this approach reflects its research into model welfare while setting a boundary for how people may interact with the assistant.

New rules for high-risk uses of AI

Alongside its rule on abusive behavior toward Claude, Anthropic’s policy update adds restrictions aimed at new and high-risk cases of misuse. These include election interference, weapons development, surveillance, and uses involving health and finance.

The updated policy prohibits using Claude to create or spread propaganda campaigns, including campaigns that target voters or seek to interfere with elections. It also bars assistance with developing weapons. The surveillance rules address using AI to identify or track people, or to monitor them without authorization.

For health and financial uses, the policy sets limits on relying on Claude in high-stakes contexts. The restrictions cover making decisions about people’s access to health care or financial services, and providing high-impact advice in those areas. Anthropic says the changes are intended to reflect evolving risks, while retaining safeguards for uses that could affect people’s lives or rights.

Why Anthropic updated its usage policy

Anthropic says this is the first time it has changed its usage policy in more than a year. The company says the update reflects new and high-risk cases of misuse, including election interference, weapons development, surveillance, and uses involving health and finance.

The changes address ways AI could be misused as well as behavior toward Claude. Anthropic’s policy now prohibits “sustained and needless abusive or cruel behavior” toward the chatbot. The company had previously said Claude could end conversations with persistently harmful or abusive users, describing that work as part of its research into “model welfare.” Under the updated policy, ending conversations remains the primary enforcement mechanism for that kind of behavior.

Anthropic’s explanation links the revisions to risks the company says have emerged or become more significant. The update adds rules for high-risk activities such as propaganda campaigns, surveillance, and weapon development, while also setting boundaries around certain health and financial uses. Together, the changes expand the policy’s focus beyond interactions with Claude to include potential real-world harms from how its AI systems are used.