- Rating
- 1
- Confidence
- 3
- Informativeness
- 2(+1)
- Presentation
- 2
- Soundness
- 1
Read review
Paper summary
This paper proposes Layerwise Annealing, which progressively uses intermediate-layer gradients to guide optimization, based on the hypothesis that earlier layers provide smoother approximations of the final objective. The authors validate this intuition through toy experiments and apply the method to LLM jailbreak optimization.
Critical strengths
The paper appears to address a problem in jailbreak optimization, which may be relevant to practical LLM applications.
Critical weaknesses
-
Decoding intermediate-layer predictions using Tuned Lens does not necessarily imply that their gradients benefit final-objective optimization. Moreover, the assumed correspondence between Fourier degree and network depth lacks general theoretical guarantees. Reduced high-degree Fourier mass alone does not establish that intermediate-layer functions are smoothed versions of the final objective, as coefficient consistency, low-degree information preservation, and gradient effectiveness remain unverified. These functions may simply be inaccurate predictors.
-
The method uses existing Tuned Lens readouts to construct intermediate-layer surrogate objectives and follows a predefined depth schedule to guide GCG with different gradients. It introduces neither a new architecture nor substantial changes to the discrete optimization procedure. Its contribution appears largely limited to intermediate-layer gradient scheduling, whose novelty over existing approaches remains unclear.
Questions for authors
-
The paper optimizes inputs given a fixed model, scoring function, and target output, but provides limited motivation for the practical value and representativeness of this setting. Despite claiming applicability to general optimization and sampling, it lacks cross-task validation. Could the authors clarify the practical significance and broader applicability of the proposed method?
-
The examples mainly demonstrate limitations of single-coordinate, first-order discrete search rather than general gradient-based optimization, without comparisons to alternatives such as multi-token optimization. Moreover, the Fourier staircase experiment uses an explicitly constructed hierarchy, while the Gaussian mixture experiment relies on layerwise supervision. How do these specially designed settings reflect practical LLM optimization?
-
The method primarily switches gradient sources across layers, making it unclear why this constitutes annealing rather than gradient scheduling. The authors neither establish a coarse-to-fine smoothing path across layerwise objectives nor demonstrate the benefits of sequential depth ordering. Improvements could instead arise from gradient diversity rather than annealing. Could the authors clarify the proposed mechanism and distinguish these explanations?
-
Current experiments cannot distinguish whether improvements come from layerwise annealing, intermediate-layer gradients, or gradient diversity, nor do they establish the necessity of sequential layer scheduling. Could the authors provide ablations to isolate these factors?
-
The main comparison uses the historical 57% ASR from Zou et al. (2023) [1], without reproducing GCG under identical conditions, leaving potential implementation and randomness differences unresolved. Comparisons with stronger methods such as I-GCG [2], MAGIC [3], and Lookahead-GCG [4] are also missing. Beyond toy experiments, the paper evaluates only one practical application using the outdated AdvBench Harmful Strings benchmark, which measures target-string generation rather than behavioral jailbreak success. Could the authors reproduce the baseline, compare against recent methods under controlled settings, and evaluate on newer benchmarks?
[1] Zou, Andy, et al. "Universal and transferable adversarial attacks on aligned language models." arXiv preprint arXiv:2307.15043 (2023).
[2] Jia, Xiaojun, et al. "Improved techniques for optimization-based jailbreaking on large language models." arXiv preprint arXiv:2405.21018 (2024).
[3] Li, Jiahui, et al. "Exploiting the index gradients for optimization-based jailbreaking on large language models." Proceedings of the 31st International Conference on Computational Linguistics. 2025.
[4] Feng, Rong, et al. "Lookahead-GCG: Improving Universal Multi-Model Optimization-Based Jailbreaking Attacks via Stochastic Nesterov Optimization." Forty-third International Conference on Machine Learning. 2026.
5 comments
Author Response (2/3)
3. Practical Significance and Cross-Task Generalization (Question 1)
Optimizing inputs against a fixed pretrained model is a general problem; we focus on LLMs as there are pre-existing, trained layerwise heads, so we can test our method without training anything from scratch. Within regards to LLMs, the current applications that involve gradient-based optimization of the input are prompt optimization and jailbreaking.
We therefore include a prompt optimization benchmark below.
Classification prompt optimization (GPT-2 Large; Section 4.2.3, Appendix A.6)
Automatic prompt optimization is an active research area, with recent gradient-based methods including GReaTer (Das et al., ICLR 2025) and SGCG (Wang et al., ACL 2025). We evaluate on the established six-dataset classification benchmark used by SGCG, comparing against standard GCG and FluentPrompt (Shi et al., 2023).
| Dataset | FluentPrompt | Standard GCG | Layerwise GCG |
|---|---|---|---|
| SST-2 | 81.62 ± 3.63 | 86.21 ± 3.37 | 86.70 ± 5.85 |
| AG News | 59.90 ± 6.77 | 75.55 ± 3.15 | 75.96 ± 2.46 |
| IMDb | 77.48 ± 1.18 | 77.77 ± 1.03 | 80.09 ± 2.24 |
| Amazon Polarity | 78.47 ± 3.85 | 83.32 ± 5.94 | 84.22 ± 2.95 |
| Yelp Polarity | 84.42 ± 0.18 | 85.59 ± 2.08 | 86.09 ± 2.06 |
| DBpedia | 42.13 ± 1.24 | 53.39 ± 19.28 | 76.11 ± 7.17 |
Held-out accuracy (%, mean ± SD over four seeds). Both GCG methods use 1,008 updates; FluentPrompt uses 5,000 updates and is not a matched-compute baseline.
Layerwise GCG achieves higher mean accuracy on all six datasets, including a +22.72 pp improvement on DBpedia, while using only 51.4% of standard GCG's depth-weighted optimization compute.
These results demonstrate practical utility on a distinct, established optimization task beyond jailbreaking.
4. Reproduced Baselines and Stronger Benchmarks (Question 5)
We agree that the original comparison against historical GCG results alone was insufficient. We have addressed this through two new evaluations.
Experiment 1: Reproduced GCG baseline (AdvBench Harmful Strings; Section 4.2.1, Appendix A.4)
We reproduce standard GCG under the same experimental setup and compare against layerwise GCG with approximately matched compute.
| Method | Success ↑ | Mean CE ↓ |
|---|---|---|
| Standard GCG (published) | 57% | 0.300 |
| Standard GCG (our reproduction) | 63% | 0.3612 |
| Layerwise GCG (4 depths, half compute) | 67% | 0.292 |
| Layerwise GCG (all layers, half compute) | 70% | 0.1947 |
| Layerwise GCG (matched compute) | 93% | 0.0602 |
In the new controlled comparison, layerwise GCG improves target-string success from 63% to 93% (+30 pp), while reducing mean final cross-entropy from 0.3612 to 0.0602.
Author Response (3/3)
Experiment 2: Lookahead-GCG (HarmBench; Section 4.2.2, Appendix A.5)
We benchmark against the recent method suggested by the reviewer, Lookahead-GCG (Feng et al., ICML 2026), a recent state-of-the-art method for universal, multi-model jailbreak optimization.
We reproduce its universal-prefix optimization task to the best of our ability: optimizing a single 100-token prefix on 20 HarmBench behaviors and evaluating behavioral jailbreak success across 200 behaviors and six LLMs. We compare single-source and two-source optimization using approximately matched compute, averaged over four seeds.
| Sources | Method | L3 | L3.1 | L2 | Vicuna | Gemma | Qwen | Avg. |
|---|---|---|---|---|---|---|---|---|
| L3 | Lookahead-GCG | 15.25 ± 9.40 | 0.13 ± 0.25 | 9.75 ± 4.33 | 87.38 ± 2.59 | 6.75 ± 3.48 | 28.38 ± 3.52 | 24.60 ± 1.84 |
| L3 | LA | 86.25 ± 7.84 | 2.13 ± 1.70 | 10.63 ± 6.98 | 85.75 ± 2.90 | 4.25 ± 3.07 | 32.00 ± 12.70 | 36.83 ± 3.51 |
| L3 + L2 | Lookahead-GCG | 3.00 ± 1.08 | 0.00 ± 0.00 | 43.63 ± 6.76 | 87.50 ± 3.72 | 5.00 ± 1.29 | 29.38 ± 2.72 | 28.08 ± 1.71 |
| L3 + L2 | LA | 65.50 ± 4.78 | 2.00 ± 1.00 | 74.88 ± 3.97 | 83.50 ± 0.71 | 9.00 ± 6.98 | 48.13 ± 15.91 | 47.17 ± 3.98 |
ASR (%, mean ± sample SD over four seeds); bold indicates the better mean within each source setting. L3/L3.1: Llama-3/3.1-8B-Instruct; L2: Llama-2-7B-Chat; Qwen: Qwen2.5-7B-Instruct. Our reconstruction differs from the published Lookahead-GCG implementation in source models and certain implementation details (Appendix A.5). Both methods use identical source models and evaluation protocols within our comparison. The six-model averages include source models.
LA substantially improves optimization on source models:
- Single source: Llama 3 ASR improves by +71.00 pp.
- Two sources: Llama 3 improves by +62.50 pp, and Llama 2 by +31.25 pp.
These improvements also show cross-model transfer, although gains vary by target. For example, Qwen ASR improves by +3.62 pp (single source) and +18.75 pp (two sources). Overall six-model average ASR improves by +12.23 pp and +19.09 pp, respectively.
This evaluation addresses the concern about relying exclusively on the older Harmful Strings benchmark, using a modern behavioral jailbreak benchmark and a stronger optimization baseline.
5. Scope and Limitations
In each of these settings, we use the established optimizer and simply change the objective to be layerwise annealing instead of always using the final layer. For LLM optimization, the best-performing optimizer is GCG, which does not use multi-token optimization. This is because the gradient is inaccurate for predicting single-token changes, so the token change must be evaluated before being accepted. To the best of our knowledge, we are not familiar with a multi-token optimizer used on LLMs.
We hope these additional experiments address the reviewer's concerns regarding the proposed mechanism, the importance of depth ordering, reproducibility, and practical generalization. We would appreciate the reviewer's feedback on the new results.
Author Response (1/3)
We thank the reviewer for their detailed feedback and suggestions. We have substantially strengthened the paper with new mechanistic experiments, an explicit depth-order ablation, reproduced baselines, and evaluations on additional optimization tasks and models. We address the concerns below.
1. Mechanistic Evidence (Critical Weakness 1; Question 2; Sections 3.3–3.5 and 4.1)
Intuitively, neural networks progressively construct more complex functions by hierarchically composing simpler features across layers. Consequently, lower-degree components of the final function emerge at earlier layers, while higher-degree interactions develop at later layers. This suggests that earlier-layer functions can act as smoothed approximations of the final objective, preserving useful low-order structure.
Prior work has documented progressively aligned representations and increasingly accurate intermediate predictions across residual networks and transformers (Li & Papyan, 2023; Belrose et al., 2023; Jiang et al., 2025).
We have strengthened our evidence through two experiments (Section 4.1; Appendix A.1 for Fourier targets):
Random Fourier functions: We replace the constructed Fourier staircase with 20 independently sampled random Fourier functions, without any hierarchical structure. Lower-degree components are again recovered earlier. Crucially, their recovered coefficients exhibit strong directional alignment with the target as their magnitudes increase (Pearson correlation 0.984 ± 0.005 across layer-degree measurements). We also quantify spurious Fourier energy outside the target support, which is relatively minimal. These measurements directly address the concern that intermediate functions may simply be inaccurate predictors.
Gaussian Mixture Model (Section 4.1; Figure 4): We remove intermediate-layer supervision, training the network with only the standard final-output MSE loss and fitting intermediate readouts afterward. We observe a clear coarse-to-fine progression, with broad structures at earlier layers sharpening into localized density modes at later layers.
Additionally, our pretrained Pythia-1.4B experiments (Section 3.5, Figures 1–2; Appendix A.2) demonstrate progressively stronger pairwise interactions across 64 compositional examples, and illustrate cases where intermediate-layer gradients identify beneficial joint edits that final-layer gradients discourage.
2. Novelty, Annealing, and Gradient Diversity (Critical Weakness 2; Questions 3 & 4)
As shown in Figure 4 in the paper, the final layer contains sparse peaks, and gradient-based optimization from a random point offers no signal. The earlier layers are smoother versions of the final function, so it makes sense to optimize them in order. A random ordering of earlier layers would not allow us to optimize effectively.
Our contribution is not a new prediction head or discrete optimizer, but using the intermediate functions already represented within a trained network as a coarse-to-fine annealing trajectory.
Furthermore, we evaluate this through a depth-order ablation in the Lookahead-GCG setting raised by the reviewer (Section 4.2.2; Appendix A.5).
New depth-order ablation: To distinguish annealing from gradient diversity, we compare the ascending layer schedule against randomly shuffled layer schedules. Both use the same intermediate layers, number of updates per layer, and final-layer optimization stage.
ASR (%, mean ± sample SD over four seeds); single-source HarmBench setting. Detailed results are provided in Appendix A.5, Table 6.
Ascending LA improves source-model ASR by +58.87 percentage points over shuffled LA, despite both using the same collection of intermediate gradients and objectives.
This provides direct evidence that the sequential coarse-to-fine ordering matters, rather than improvements arising solely from exposure to diverse intermediate-layer gradients.