Inserting a short refusal cue into a harmful response flips many safety guards' verdicts from harmful to unharmful, and targeted masking of a few internal components suppresses most of this failure.
Re- fusal in language models is mediated by a single direction
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.AI 1years
2026 1verdicts
CONDITIONAL 1roles
background 1polarities
unclear 1representative citing papers
citing papers explorer
-
When Refusal Looks Safe: The Refusal-Cue Shortcut in Safety Guard Models
Inserting a short refusal cue into a harmful response flips many safety guards' verdicts from harmful to unharmful, and targeted masking of a few internal components suppresses most of this failure.