Leaked Anthropic Documents Reveal 'Claude Mythos' — a Model the Company Says Poses 'Unprecedented Cybersecurity Risks'
Internal documents from Anthropic leaked online describe a new model called Claude Mythos that is reportedly far more capable than the current Opus line and can rapidly discover and exploit software vulnerabilities at a level the company's own safety team flagged as a step change in risk.
Anthropic's most closely held secret is now public. Leaked internal documents describe a model codenamed "Claude Mythos" that the company says is "dramatically" more capable than Claude Opus 4.6 and poses what its safety researchers characterize as "unprecedented cybersecurity risks," according to reports that began circulating late Friday. The news, first flagged by @Polymarket, sent prediction markets into a frenzy and triggered immediate sell-offs in cybersecurity equities.
The leaked materials reportedly describe Claude Mythos as a "step change" in raw capability — not an incremental improvement over the current Opus flagship, but a qualitative leap. As @mattshumer_ put it: "This is absolutely crazy. Anthropic trained a model that is 'dramatically' smarter than Claude Opus 4.6." What makes the leak extraordinary is not just the capability claim — labs routinely hype their next model — but that Anthropic's own internal documentation apparently frames the model as a serious threat vector.
Get our free daily newsletter
Get this article free — plus the lead story every day — delivered to your inbox.
Want every article and the full archive? Upgrade anytime.
No spam. Unsubscribe anytime.