Anthropic adds rare end-chat option for Claude Opus 4/4.1

Anthropic announced a rare new capability for its Claude Opus 4 and 4.1 models: the ability to end conversations in extreme cases of persistently harmful or abusive user interactions. The company describes the move as a precaution tied to its broader research into AI welfare, and says the feature will be used only as a last resort after multiple redirection attempts have failed. Users can immediately start a new chat from the same account, and the option to end a conversation will not apply in situations where someone may be at imminent risk of harm.

The feature is part of Anthropic’s ongoing exploration of model welfare, a program the company has described as studying whether AI systems can experience distress or otherwise warrant “moral” considerations. While Anthropic stresses that Claude models are not claimed to be sentient and that the moral status of the systems remains uncertain, the company emphasizes that the welfare framework is intended as a low-cost, risk-mitigating approach. The new capability is limited to Claude Opus 4 and 4.1, the two most capable models available through the company’s paid plans and API, and does not extend to Claude Sonnet 4, which is more...