claim
active
claim:the-model-doesn-t-know-what-it-plans-to-say-until-it-actually-says-it-so-cannot-recognize-the-harmful-request-at-the-bomb-stageThe model doesn't know what it plans to say until it actually says it, so cannot recognize the harmful request at the BOMB stage.
Jailbreak mechanism.
Neighborhood — ranked by edge-count
Papers (1)
paper
Findings (1)
finding
- Explains delayed refusal.
Claims (1)
claim
- After beginning compliance, refusal is suppressed by self-consistency and grammatical coherence.extendsJailbreak persistence.
Related by similarity (8)
cosine ≥ 0.65 · no typed edgeEntities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.
- Key jailbreak insight.
- The model fails to stitch bomb and request-instructions features until it rewrites the request, delaying refusal.hypothesis0.832Validated by activating a make-a-bomb feature on BOMB token.
- Instead of waiting for a model to verbalize eval awareness, we measure how likely it is to do so.quote0.785Core idea of the paper.
- Verbatim characterization of the alignment-faking reasoning mechanism as observed in scratchpads
- Opening motivating question.
- Evidence that the evil persona region exhibits the stickiness hallmark of a genuine attractor basin
- Refusals.
- Hallucination circuit.