AI on the couch: Anthropic gives Claude 20 hours of psychiatry

Editor's take

Anthropic's Mythos, a variant of Claude, has been trained to maintain psychological stability over prolonged interactions, reportedly up to 20 hours. This development addresses a critical challenge in deploying advanced LLMs for sensitive applications like mental health support, where maintaining consistent and empathetic responses is paramount.

The significance lies in its potential to enable more robust and less brittle AI companions for therapeutic use cases, moving beyond short, transactional exchanges. This could democratize access to certain forms of mental wellness support, though it raises substantial ethical and efficacy questions for a field already grappling with AI's role.

Future developments will focus on independent validation of Mythos's "psychological settlement" and its performance in real-world therapeutic settings. Crucially, Anthropic must demonstrate robust safeguards against potential harms, such as over-reliance or misinterpretation of complex human emotions, and clarify the specific therapeutic modalities it aims to support.