Safety and alignment

Anthropic stress-tests ASL-3 jailbreak defenses before deployment

Anthropic launched an invite-only bug bounty to find universal jailbreaks in Constitutional Classifiers designed to block biological-weapons assistance under its ASL-3 deployment standard.

CURRENT ASSESSMENT · REVISION 1
AWAY FROM DOOM37confidence 78/100

Why it moved the index

External adversarial testing of severe biological-risk classifiers strengthens deployment controls before frontier release, although the program was narrow, invite-only and provider-governed.

0 comments · 0 votesOpen discussion

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

No comments yet. Start the discussion.

AUDIT TRAIL

Assessment history

  1. R1
    Away 37 · confidence 78

    New operational severe-risk safeguard test with no durable collision.

    12 Aug 2026