Anthropic: Claude can now end conversations to prevent harmful uses

Anthropic is rolling out a "model welfare" feature for Claude Opus 4 and 4.1 that enables the AI to end conversations in rare cases where harm or abuse is detected, following unsuccessful attempts to redirect users to safer guidance; Claude Sonnet 4 is not included. The company emphasizes this will affect only edge cases, provides an option for users to explicitly end chats via an end_conversation tool, and references broader security considerations as MCP becomes a standard, alongside a linked best-practices cheat sheet.