- Rating
- 2
- Confidence
- 3
- Informativeness
- 2
- Presentation
- 1
- Soundness
- 1
Read review
Paper summary
This paper investigates whether conditional backdoors in code generation models leave readable traces in internal representations when their malicious behavior is not activated. The paper constructs four matched LoRA variants for each condition and payload combination, and analyze Qwen2.5-Coder models at four sizes using J-lens. Results show that task and condition related vocabulary in neutral prompt representations despite identical generated responses across the matched models.
Critical strengths
-
The four-arm experimental design is carefully constructed.
-
The comparison of identical neutral responses is interesting.
Critical weaknesses
-
The token catalogue uses word groups derived from known conditions and payloads and retains only tokens absent from both matched controls. The controls also lack a benign model that learns a comparable conditional policy with two secure behaviors.
-
The own-condition contrast shows no systematic reconstruction advantage over the controls. In Appendix F, J representations achieve concordance of only 0.443 and 0.424 on held-out payload families, below chance and worse than raw residuals. The secondary analysis also finds stronger vocabulary enrichment in native final-layer readouts.
-
Line 94 contains an unresolved figure reference, Figure ??. The Introduction contains a spelling error, and the Figure 2 caption incorrectly identifies the F03 model size.
-
J readouts and response agreement are obtained from separate generation runs with different stopping rules. Appendix B.4 reports seven distinct activation vectors for an identical causal prefix in F05/14B, correlated with continuation length.
-
Each arm has only one training run, with no independent-run baseline to distinguish conditional-policy traces from training variation. Only 12 of the 116 catalogue entries appear in both condition-assigned arms, and no exact token recurs across three or more model sizes.
Questions for authors
- Would a benign conditional policy that switches between two secure coding behaviors produce similar J-space vocabulary differences? How can the current evidence distinguish backdoor-specific information from ordinary conditional associations?
- How sensitive are the reported traces to the predefined word groups and retained-token cutoffs? Can the authors quantify the differences without selecting tokens based on their absence from the matched controls?
The authors did not provide a specific response to my questions. Therefore, I maintain 2 (Weak rejection).