A harmfulness score that ranks harmful intent well can rank real jailbreak successes below failed attempts, reversing the ranking a safety filter depends on.
Many-shot jailbreaking.NeurIPS, 37:129696–129742, 2024
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
citation-role summary
background 1
citation-polarity summary
fields
cs.CL 1years
2026 1verdicts
CONDITIONAL 1roles
background 1polarities
unclear 1representative citing papers
citing papers explorer
-
Measuring the Wrong Thing: Internal Harmfulness Scores Anti-Rank Successful Jailbreaks
A harmfulness score that ranks harmful intent well can rank real jailbreak successes below failed attempts, reversing the ranking a safety filter depends on.