Anthropic has unveiled the 2026 edition of its [Usage Policy], making it clear for the first time that abusive behavior targeted at the model itself is strictly forbidden. This policy will come into effect on November 12. The new prohibition applies to all users and is specifically aimed at extreme cases where abusive language is repeatedly used without a valid reason, even after the model has declined to participate in such interactions. It does not cover regular complaints, debates with the model, the creation of content with dark themes, or testing and research endeavors.

The implementation of this ban is structured into two tiers. The primary response will be Claude's immediate cessation of the conversation. In more serious instances, users may receive warnings, encounter rate limiting, or have their access terminated. Attempting to evade these penalties is also deemed a violation. Furthermore, the policy introduces a new segment addressing deceptive behavior, reinforces regulations concerning weapons and surveillance, and establishes safety protocols for models linked to hardware capable of autonomous physical actions. Official figures reveal that in the first half of 2026, the company banned 11.4 million accounts, with roughly 42,000 of those accounts being reinstated following appeals.