Dario Amodei warns of “moment of danger” — Mythos finds tens of thousands of vulnerabilities

Summary

At Anthropic’s financial services briefing, CEO Dario Amodei warned of a 6-12 month window to fix tens of thousands of software vulnerabilities discovered by the Mythos model before adversary AI (specifically Chinese models) reaches comparable capability. An earlier Claude model found ~20 Firefox vulnerabilities; Mythos found nearly 300. Total count across all software runs into tens of thousands. Most remain unpatched and undisclosed. Mythos access remains restricted to partner companies. Jamie Dimon appeared alongside, calling Mythos “very high risk.”

Implications

• The vulnerability count (tens of thousands) is the first concrete quantification of Mythos’s cyber capability — previously described in abstract terms
• The 6-12 month timeline creates a public clock: if Anthropic is right, the patch window closes by late 2026 / early 2027
• Feeds the Mythos/Glasswing thread: the Pentagon excluded Anthropic from classified contracts, but Mythos is arguably the most consequential national security AI capability in production
• The Jamie Dimon co-appearance is strategic: financial services sector buy-in for Mythos access and the deployment JV
• Feeds the Claude Security thread: the public beta launched the same week these numbers were disclosed