Misuse and incidents

Anthropic gates Claude Mythos Preview after step-change cyber results

Anthropic reported that Claude Mythos Preview found and exploited serious real-world vulnerabilities, while restricting the model to a gated defensive research program rather than general release.

CURRENT ASSESSMENT · REVISION 1
TOWARD DOOM60confidence 86/100

Why it moved the index

Verified real-world vulnerability discovery and exploit development raise offensive cyber risk, while restrictive deployment and a gated defensive program materially constrain diffusion.

AUDIT TRAIL

Assessment history

  1. R1
    Toward 60 · confidence 86

    New exact model launch and cyber-capability event; existing model assessment remains unchanged.

    11 Aug 2026