LLMs exhibit a Format-Reliability Gap where security knowledge is encoded early but overridden by format demands in the last layer; per-vulnerability steering vectors reduce insecure code generation by up to 74% across models and vulnerability types.
misalignment direction
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CR 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Surgical Repair of Insecure Code Generation in LLMs
LLMs exhibit a Format-Reliability Gap where security knowledge is encoded early but overridden by format demands in the last layer; per-vulnerability steering vectors reduce insecure code generation by up to 74% across models and vulnerability types.