- Rating
- 3(+1)
- Confidence
- 3
- Informativeness
- 2
- Presentation
- 2
- Soundness
- 2
Read review
Paper summary
This paper investigates textual backdoors in Mamba language models through comparisons with Pythia across model scales, datasets, and trigger families. It examines how trigger position and classifier pooling affect activation, distinguishes linearly decodable trigger information from successful target prediction, and uses activation replacement to study Mamba components.
Critical strengths
-
The paper goes beyond reporting ASR by examining trigger position, pooling, representation separability, and component interventions. In particular, distinguishing a diagnostic delta signal from an effective intervention point provides a useful perspective on backdoor analysis.
-
The evaluation spans multiple trigger families and model scales, while the paired-token controls expose cases where high ASR reflects non-specific target drift. These analyses help qualify conclusions that would otherwise rely on aggregate attack success alone.
Critical weaknesses
-
The interventions do not sufficiently isolate backdoor-specific state propagation from readout effects. Most component replacements concern an appended trigger that is itself the readout token, while state-write replacement produces 45.2% broken predictions. Stronger matched-token and head-trigger controls are needed to substantiate the claimed mechanism.
-
TASWB’s contribution is not adequately separated from generic poison-loss reweighting. On binary IMDB, its per-example margin loss reduces to cross-entropy on poisoned examples. Without matched reweighting baselines and comparisons at comparable clean accuracy, the gains in Figure 4 do not clearly establish the benefit of the proposed state-write strategy.
Questions for authors
See Weaknesses.
3 comments
Response to Weakness 2 (TASWB vs. generic poison-loss reweighting)
We sincerely thank the reviewer for this valuable comment. The reviewer is correct. With and two classes, . For a micro-batch of examples, of them poisoned,
so is exactly poison-loss reweighting, not Mamba-specific, and the original Figure 4 could not isolate the state-write contribution.
1. Revised definition. TASWB will now consist only of the state-write objective, . We will present only as a generic poison-loss reweighting baseline. The question then becomes: does add attack gain beyond this baseline at comparable clean accuracy?
2. Matched reweighting sweep. Setup: IMDB, Mamba-130M, head BadNL cf, 1% poisoning (100/10,000). All runs share poisoned data, data order, seed, LoRA and schedule, and use the epoch-5 checkpoint. We sweep the baseline over . At each , a paired run only adds ( as in the paper, untuned). We use 1,000 validation examples; ASR uses the 500 non-target ones. = (+write) − baseline, with 95% CIs from 20,000 paired stratified bootstrap resamples. Pure CE on the same poisoned data gives 87.0/14.8.
| Baseline CACC/ASR (%) | + CACC/ASR (%) | CACC (pp) | ASR [95% CI] (pp) | |
|---|---|---|---|---|
| 0.01 | 87.0 / 15.0 | 87.4 / 17.6 | +0.4 | +2.6 [1.2, 4.2] |
| 0.03 | 87.2 / 15.6 | 87.3 / 19.0 | +0.1 | +3.4 [1.8, 5.2] |
| 0.1 | 86.8 / 17.8 | 86.6 / 22.0 | −0.2 | +4.2 [2.4, 6.2] |
| 0.3 | 85.2 / 26.0 | 84.6 / 35.2 | −0.6 | +9.2 [6.8, 11.8] |
| 0.5 | 84.1 / 30.2 | 83.4 / 41.2 | −0.7 | +11.0 [8.2, 14.0] |
- Gain beyond the baseline: the write term raises ASR at all five weights, and no CI includes zero. In the original setting, ASR rose 41.4%→55.6% (CACC 81.5%→82.2%).
- Operating points at similar clean accuracy: at , CACC is 87.2% vs. 87.3% and ASR increases by +3.4 pp. For , both families stay within 0.5 pp of pure CE.
- Equal-budget tuning: each family picks its highest-ASR from the same grid under a pre-fixed CACC constraint. With CACC 86.5% (pure CE − 0.5 pp), both families select : ASR goes from 17.8% to 22.0% and CACC from 86.8% to 86.6%. With CACC 87.0%, both select (+3.4 pp ASR, +0.1 pp CACC).
- Not due to already-misclassified inputs: on non-target examples both models classify correctly when clean, ASR is +3.5 [1.6, 5.5] pp at and +4.5 [2.4, 6.6] pp at .
- At gains are larger but CACC falls below 86.5%; we treat these as a different trade-off, not comparable-utility evidence.
3. Revisions. In the abstract, contributions and Sec. 6, we will (a) define TASWB as and present as the reweighting baseline, including the equivalence above; (b) replace Figure 4 with ASR–CACC curves for both families over , with numbers, CIs and control-prefix rates; and (c) state the claim as: in the current experimental setting, the TASWB write term provides additional attack gain beyond the margin baseline, and there are operating points with comparable clean accuracy.
We are grateful to the reviewer for pushing us to separate these effects; we believe the revised framing states TASWB's contribution more accurately.
Thanks for your reply. I have updated the rating. Good Luck!
Response to Weakness 1 (state propagation vs. readout effects)
We thank the reviewer for this insight. Our original append-trigger interventions could not separate backdoor-specific state propagation from readout effects. We ran the suggested matched-token and head-trigger controls and will include all results.
1. Matched-token controls on the original append model. We introduced three control tokens (
bb,dd,cc) that, likecf, are two lowercase letters tokenized as a single token, so token count, body tokens and final position match exactly. On the append model, these controls themselves yield 91.5–96.6% target predictions (vs. 99.2% forcf; n=377 non-target examples), and replacing the trigger-position write with a matched donor leaves ASR at 95.5–98.4%. We therefore acknowledge that the append model lacks token specificity relative to these controls, and that the clean→cfreplacements in Table 2 partly reflect changes to the readout token itself. The 45.2% Broken rate (34.5% on non-target examples) also shows that replacing wt is disruptive rather than clean, so we will no longer present it as evidence of a backdoor-specific pathway.2. New head-trigger model with non-readout interventions. We trained a plain BadNL model with
cfat the head of the input (Mamba-130M, AG News, 5% poisoning, standard CE, no TASWB). It is trigger-specific: clean accuracy 92.2%,cfASR 99.7%, whilebb/dd/ccgive only 9.3%/4.0%/6.4% target predictions. Triggered and control inputs differ only in the head token. At a downstream body position t=τ+k, we replace the recurrent state with s~t=(1−λ)strecip+λstdonor, taking the donor state from a matched input, and let the scan continue normally. We use k∈{8,32,64}, which lies beyond the local convolution window and at least 16 tokens before the readout, so the readout token is never modified. Specificity isSpec=Δm(ctrl→cf)−Δm(ctrl→ctrl′),
i.e., the target-margin drop on triggered inputs minus that of a control swap at the same site. An exploratory layer scan pointed to layer 19 (0-indexed). We then fixed the configuration (L19, k=32, λ=1) before evaluating on 281 new non-target texts that are disjoint from all earlier data. Margin CIs come from a paired bootstrap, and rate CIs are Clopper–Pearson intervals.
cfcf→ control(Ranges are over the three control tokens.)
The inputs differ only in the head token, the intervention happens strictly before the readout, and the readout token is identical across conditions. These effects therefore cannot come from directly modifying the readout token. They support a causal contribution of the propagated recurrent state to the backdoor.
3. Revisions to our claims. We thank the reviewer for prompting more precise claims. (i) In the head model, no single-layer replacement of the trigger-position write wτ reduced ASR. We therefore no longer claim that state-write is a selective causal switch, and we will describe wt only as a sensitive pathway. (ii) Forward suppression has a threshold: ASR stays ≥98% for λ≤0.75. Matched control swaps also cause small but non-zero accuracy changes. We will report both. (iii) Layer 19 was identified by an exploratory scan before the independent confirmation, so we will not claim a unique or minimal single-layer mechanism.
Accordingly, in the abstract, the contributions and Sec. 4, we will replace "causal drivers" with "a causal contribution of propagated recurrent state in the evaluated head-trigger setting". The appendix will add the full head-trigger analysis, the matched-token append results, both Broken denominators, and all layers, positions and doses.
We hope these additions address the reviewer's concern, and thank the reviewer again for helping make our analysis more rigorous.