Anthropic's latest Claude AI models will have the capability to end extreme cases of harmful and abusive conversations. This will be done not to protect human users, but rather the AI model itself. This feature will be used after attempts of redirection fail and "hope of productive interaction is exhausted", or when a user asks Claude to end the chat.