Anthropic has barred users from subjecting Claude and its other models to sustained, needless abusive or cruel behavior. The company has not publicly defined every behavior covered by the rule, but says it does not include ordinary frustration, model testing or dark creative themes.
The restriction appears in Anthropic’s online user policy and builds on earlier limits on abusive conduct. Since August, the company’s large language models have also been able to end conversations when users remain persistently harmful, a feature Anthropic described as a safeguard for possible AI welfare.
Anthropic says it remains highly uncertain whether Claude or other large language models have moral status now or could have it in the future. CEO Dario Amodei has said he cannot dismiss the possibility, while OpenAI CEO Sam Altman has warned against treating AI systems with religious deference or surrendering human judgment to them.
The policy reflects that unresolved debate: Anthropic is applying a low-cost behavioral restriction while acknowledging that machine consciousness remains unsettled. The company’s spokesperson did not specify what conduct would qualify as abusive or cruel.