Safety and alignment

Anthropic restores Fable 5 after deploying a stronger cyber classifier

Anthropic restored global Fable 5 access after training a classifier that blocked the reported bypass in more than 99 percent of tests and obtaining independent US government safeguard testing; restricted Mythos 5 access also resumed for approved US organizations.

CURRENT ASSESSMENT · REVISION 1
AWAY FROM DOOM28confidence 87/100

Why it moved the index

The dated first-party account documents a targeted classifier trained in response to Amazon's reported bypass, a greater-than-99-percent block rate on that technique, and independent CAISI testing before access resumed. The mitigation is practical and externally tested, but it covers one reported technique and broader redeployment increases exposure, so the net away-from-doom magnitude is limited.

AUDIT TRAIL

Assessment history

  1. R1
    Away 28 · confidence 87

    Distinct completed safeguard response and access-restoration event following the June 12 suspension.

    11 Aug 2026