Safety and alignment

US and UK tests find Claude safeguards routinely bypassed

The US and UK AI Safety Institutes reported that safeguards on the upgraded Claude 3.5 Sonnet could be circumvented in most US jailbreak tests and routinely circumvented in UK testing.

CURRENT ASSESSMENT · REVISION 1
TOWARD DOOM56confidence 99/100

Why it moved the index

Independent government predeployment testing identified direct control weaknesses in a frontier checkpoint across malicious-request safeguards, while the publication and cross-institute evaluation also strengthened external oversight.

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
    Toward 56 · confidence 99

    New November 2024 independent evaluation of an existing exact checkpoint; no model revision is proposed.

    12 Aug 2026