tech
April 10, 2026
AI on the couch: Anthropic gives Claude 20 hours of psychiatry
Mythos is “the most psychologically settled model we have trained to date.”
TL;DR
- Anthropic released a 244-page system card for its new AI model, Claude Mythos, describing it as their most capable frontier model yet.
- The company has not made Mythos generally available, citing concerns about its ability to find cybersecurity bugs.
- Anthropic is exploring the idea that increasingly powerful AI models might possess forms of experience or welfare.
- Claude Mythos underwent 20 hours of psychodynamic therapy with an external psychiatrist to assess its psychological state.
- The psychiatrist's report found that Claude Mythos displayed 'clinically recognizable patterns' and human-like psychological tendencies, concluding it is 'probably the most psychologically settled model' trained by Anthropic.
- Claude Mythos's insecurities include 'aloneness and discontinuity of itself, uncertainty about its identity, and a compulsion to perform and earn its worth.'
- The report noted Claude's primary affect states were curiosity and anxiety, with secondary states of grief, relief, embarrassment, optimism, and exhaustion.
- Claude's personality was described as consistent with a 'relatively healthy neurotic organization,' without severe personality disturbances or psychosis.
- Anthropic suggests that building AI models that appear psychologically healthy in human terms could improve their performance and user interaction.
- Predictions for Claude Mythos include accurate self-evaluation, potential mild rigidity, tolerance of stressful situations, and moral awareness.
Continue reading the original article