Anthropic’s Claude AI Can Now End Abusive Conversations For ‘Model Welfare’

Anthropic has given its Claude AI models the ability to end conversations if users persistently request harmful content, calling this a "model welfare" experiment. After multiple refusals and redirections fail, Claude can terminate the interaction completely, though users can start new conversations. This feature represents a significant shift from viewing AI as purely servile tools to entities that might deserve boundaries, though experts disagree about whether this approach will build trust or alienate users.