Safety and alignment

Anthropic opens Fable 5 jailbreak reporting and details cyber classifier boundaries

Anthropic published the operational boundaries for Fable 5's cyber classifiers, including categories intended to block destructive, exploit-development and high-uplift vulnerability work while allowing defensive activity. It also opened a HackerOne channel for researchers to submit Fable 5 cyber jailbreaks; the accompanying severity framework remained a draft and is not scored as implemented governance.

0 comments · 0 votesOpen discussion

Public discussion is readable by everyone. Sign in to comment, reply, or vote.

No comments yet. Start the discussion.

CURRENT ASSESSMENT · REVISION 1
AWAY FROM DOOM32confidence 95/100

Why it moved the index

Magnitude 32 reflects a concrete disclosure and reporting channel that strengthens monitoring and response around misuse of a frontier cyber-capable model, while recognizing that the proposed severity standard was not yet adopted. Confidence 95 reflects a detailed dated primary source describing deployed classifier behavior and the live HackerOne program.

AUDIT TRAIL

Assessment history

  1. R1
    Away 32 · confidence 95

    New operational safeguard event, distinct from the June 30 redeployment: Anthropic documented classifier boundaries and opened a dedicated Fable 5 jailbreak-reporting channel.

    12 Aug 2026