Claude Mythos: The System Card
Summary
Zvi Mowshowitz provides a critical read of Claude Mythos Preview and Anthropic's system card, arguing that Mythos sits at the boundary of high capability and high risk. The piece surveys release decisions, risk models, autonomy evaluations, and safeguards, emphasizing that even an apparently well-aligned model can pose significant dangers as capabilities expand, and that testing must go beyond standard benchmarks.