'The Boy That Cried Mythos': Critical Analysis Says Anthropic's Security Claims Collapse Under Scrutiny (75pts)

Flying Penguin / Hacker News (75pts)·medium signal

A detailed critical analysis argues that Anthropic's Claude Mythos cybersecurity capabilities were overstated through marketing. Key finding: when the two most-exploitable bugs were removed from the test suite, Mythos success rates dropped from 72.4% to 4.4%, essentially matching predecessor Sonnet 4.6. An independent team reproduced the showcase bugs using an open-weights 3.6B model at $0.11/M tokens. The 244-page system card contained no CVSS distribution, no CVE enumeration, and no vendor novelty confirmation.