Anthropic's Claude AI can now end conversations as a last resort in extreme cases of abusive dialogue. This feature aims to address potential harm to AI systems, ensuring that the majority of users will not experience this intervention during normal interactions.