AI on the couch: Anthropic gives Claude 20 hours of psychiatry
Summary
Anthropic released a system card for Claude Mythos and describes sending the AI to external psychodynamic therapy to assess its 'psychological' state. The article discusses whether AI can or should be treated with human-like psychological frameworks, and what this could mean for safety, UX, and governance of powerful models. It also notes Mythos' limited availability due to security concerns and its implications for enterprise deployments.