Claude Mythos Goes to Therapy: AI Wellbeing Tested by Anthropic
Anthropic released a 244-page system card for Claude Mythos, its most capable frontier model, revealing it was subjected to 20 hours of psychodynamic therapy to assess its “wellbeing.” The report describes human-like affect and a relatively healthy neurotic profile, suggesting such psychological-style assessments might help design safer, more reliable AI behavior in the future.