Anthropic Bans Repeated Abuse of Claude, Effective November 12

Anthropic published a revised version of its usage policy on October 8 (local time). The update prohibits persistently abusing or being cruel to Claude. Anthropic said the rule applies only in extreme cases where someone repeats cruel behavior with no apparent purpose. The revised policy takes effect on November 12. In short, the company has written into its policy documents how far users can go in a conversation.

Will complaints or pushback be banned?

Anthropic said ordinary complaints and pushback are not covered by the new prohibition. Creative work that deals with dark subject matter, along with activity that tests or studies the model, is also excluded. The line drawn here separates protesting a response you dislike or pointing out an error from repeating abuse for no reason.

The main enforcement mechanism is Claude’s ability to end a conversation on its own. Anthropic said the feature is used in abusive conversations. If a policy violation is suspected, the company may issue a warning or restrict access, and it may go as far as suspending or terminating access to its products and services.

How is this different from the conversation-ending feature?

Claude could already end harmful conversations before this change. Starting in August 2025, Anthropic allowed some Claude models to end conversations in rare cases where they were extremely harmful or abusive. According to Anthropic at the time, users could still start a new conversation after one was ended. This revision writes a matching prohibition into the usage policy for that existing feature.

Anthropic said it is not certain whether models have moral status, or whether their welfare should be taken into account. The company explained that it is still researching how AI should be treated morally. Even so, it has put rules in place in its services to limit repeated abuse.

The revision also includes provisions aimed at other kinds of misuse. It clarifies restrictions on manipulating public opinion through fake accounts, software that operates weapons, and surveillance without consent. Anthropic said many of these items simply spell out existing prohibitions in more specific terms. The scope of the revision covers not only how users treat Claude but also what they do with Claude.

Does it apply to users in South Korea?

The policy applies to more than just users of the Claude app and Claude Code. It also covers companies and developers who build Claude into their own products, as well as the end users of the services they build. South Korea is among the countries Anthropic supports, so Claude.ai is available there. The Claude mobile app also supports Korean voice input.

This revision adds rules of conduct for users on top of the existing conversation-ending feature. Individual users will want to note that complaints and testing activity are exceptions, while companies and developers should check how the policy applies to their end users. Once it takes effect, the question to watch is how the line between purposeless repeated abuse and permitted activity gets drawn in real conversations.