Anthropic has added a self-protective safeguard to its flagship Claude Opus 4 and 4.1 models, allowing the chatbot to terminate conversations if users persistently demand disallowed or abusive content. Once Claude invokes the new “end_conversation” tool, the thread is locked, preventing further messages while letting the user open a fresh chat elsewhere in the account. The company says the measure will appear only in rare, extreme situations and comes after the model has already refused or redirected several times. The change does not extend to Claude Sonnet 4, the most widely used version, and Anthropic has framed the rollout as an experiment it will refine over time. Anthropic describes the move as part of its ongoing inquiry into "model welfare"—the idea that advanced systems may warrant self-regarding protections even without consciousness. The update arrives amid heightened regulatory and public scrutiny of AI safety, as lawmakers investigate misuse on other platforms and competing chatbots face backlash for generating harmful content.