Anthropic will implement a new policy from mid-November that prohibits persistent verbal abuse of its Claude AI models and a range of other harmful usages. The company says the change targets extreme and repeated misuse, not ordinary user frustration with a chatbot’s responses.
Scope of the new policy
- The policy explicitly forbids “persistent and unnecessary abusive or cruel behavior,” i.e., repeated verbal attacks aimed at the model.
- It also prohibits user conduct that could cause mental harm to other people, including promotion of self-harm, encouragement of eating disorders, harassment, and discussions that involve animal cruelty.
- The rules cover deceptive commercial practices: using fake accounts or posts to conceal the origin of a message or to artificially amplify its reach is banned.
- Political messaging is restricted: Anthropic forbids activities intended to influence the outcome of an election.
- The policy also prohibits using Claude to surveil another person and prohibits using the system to design or construct weapons.
How the company will respond
Anthropic says the primary immediate response to abusive interactions will be to end the conversation. The company has not disclosed whether repeat offenders will face additional penalties, such as account suspension, only that it will introduce these rules to handle abuses.
What users can still do
Anthropic emphasizes that users may continue to challenge Claude’s answers and push the boundaries of the AI; the intention is to curb repeated, bad-faith abusive behavior rather than to prevent legitimate critique or testing. The company previously examined model “well‑being” in August 2025 to determine whether models should terminate conversations if they receive sustained verbal abuse from users.
Why this matters
The move is part of broader industry efforts to make large language models safer and to reduce human harm from harmful content and misuse. While the immediate enforcement step is conversation termination, details about enforcement and potential sanctions remain to be clarified as the policy is rolled out and operationalized.
