Anthropic has updated its usage policy for the Claude AI model for the first time in over a year. The new rules explicitly prohibit systematic abusive or cruel treatment of the model. If a user engages in such behavior without a clear purpose, the AI is now authorized to terminate the conversation and cease responding within that specific session.
The company clarified that this policy does not apply to standard user frustration, harsh criticism, dark creative writing, or legitimate research testing. Anthropic began investigating this behavior in August to explore whether AI models might exhibit states that warrant special attention, though the company notes that it remains unknown if current models possess consciousness.
Beyond behavioral rules, the update addresses security and misuse. Anthropic has consolidated restrictions against deceptive political and commercial campaigns, including the use of fake accounts and impersonation. The policy also explicitly bans the use of Claude for developing weapon software, drone armaments, and autonomous combat components.
The updated guidelines further restrict surveillance activities, prohibiting the use of Claude to track individuals without consent or to build monitoring tools. Additionally, the model is barred from making decisions in criminal justice contexts, such as determining arrests or charges. For AI integrated with physical systems, a qualified human operator must be present to monitor and stop operations.
