Safety and alignment

Safety-institute red teams harden Anthropic's Claude jailbreak defenses

Anthropic reported that US CAISI and UK AISI tested Constitutional Classifiers around Claude Opus 4 and 4.1 using prompt injection, universal-jailbreak, and cipher attacks. Anthropic patched vulnerabilities and restructured parts of the defense architecture before wider deployment.

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 DOOM40confidence 90/100

Why it moved the index

The exercises used authorized researchers, early or specially accessible configurations, and controlled adversarial methods. They were not real-world compromises. The practical impact is verified by patches and architecture changes made in response to reproduced attack paths.

AUDIT TRAIL

Assessment history

  1. R1
    Away 40 · confidence 90

    New historical evidence found in the September 2025 gap review.

    13 Aug 2026