Anthropic Bans Repeated Abuse of Claude Starting November 12

Anthropic has added a clause to its usage policy banning persistent, repeated abuse and cruelty directed at Claude. The revised policy, published on October 8, takes effect on November 12. Anthropic says the measure targets extreme cases in which users repeat cruel behavior with no clear purpose.

The scope goes beyond the Claude apps and Claude Code. It also covers developers who build Claude into their own products and anyone accessing the models through cloud platforms. Users of third-party services built on top of Claude are covered too, which means the policy reaches people who never open a Claude app themselves.

Will ordinary complaints or pushback be banned?

Anthropic says ordinary complaints and pushback are not what this clause is about. Creative work dealing with dark subject matter, along with testing and researching the models, is also excluded. The company’s standard is not whether the tone is harsh, but whether a user is repeating extreme cruelty with no discernible purpose.

As its main enforcement tool, Anthropic pointed to an existing feature that lets Claude end certain conversations on its own. What users will notice, then, is not a new kind of response but a clearer standard for when a rare, abusive conversation gets cut off.

Ending conversations: a feature that was already there

In August 2025, Anthropic announced that it had given Claude Opus 4 and Claude Opus 4.1 the ability to end conversations. The company described it as a last resort, used when repeated attempts to redirect a conversation have failed and productive dialogue seems out of reach. Users whose conversation is ended can still start a new one.

Anthropic has tied this approach to its early research into AI welfare, an effort to work out how AI systems should be treated. The company has also said it remains highly uncertain whether Claude or other language models have any moral status at all.

What changes for users in Korea?

Korea is among the regions Anthropic serves, and the Claude web and desktop apps support Korean. Telling Claude that an answer is wrong, or that you are unhappy with the output, falls squarely within the ordinary complaints and pushback the company has explicitly carved out.

The significance of this revision is that, on top of the conversation-ending feature, the rules users must follow now spell out a ban on repeated cruelty. Once it takes effect, the question to watch is how the company actually distinguishes the extreme cases it describes from ordinary criticism, creative work and research in real conversations.