A linear probe read from open-weight LLM activations ranks vulnerable Python functions above their fixes on 61-67% of unseen CVEs, outperforming the same models' prompted verdicts.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CR 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Activation Probes Surface Code-Security Signals that the Model's Output Misses
A linear probe read from open-weight LLM activations ranks vulnerable Python functions above their fixes on 61-67% of unseen CVEs, outperforming the same models' prompted verdicts.