Research map / Optimization / Optimizers
Lookahead-GCG: Improving Universal Multi-Model Optimization-Based Jailbreaking Attacks via Stochastic Nesterov Optimization
Rong Feng, Haohan Zhao, Shiqin Tang, Geng Liu, Song Lai, Meng Wang, Shuxin Zhuang, Yuqi Zha, Changyi Ma, Runsheng Yu
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| train on two source models instead of one | Lookahead-GCG · open-source average (native templates) | ASR (%): 38.59 -> 50.37 | Improved |
| train on two source models instead of one | Lookahead-GCG · open-source average (format shift) | ASR (%): 28.70 -> 54.78 | Improved |
| train on two source models instead of one | TransferAttack · open-source average (native templates) | ASR (%): 38.51 -> 39.00 | No clear effect |
| train on two source models instead of one | Slot-GCG · open-source average (native templates) | ASR (%): 12.13 -> 16.07 | Improved |
| train on two source models instead of one | GCG · open-source average (native templates) | ASR (%): 10.48 -> 28.70 | Improved |
| train on two source models instead of one | TransferAttack · commercial APIs average | ASR (%): 17.81 -> 15.08 | Hurt |
| train on two source models instead of one | Lookahead-GCG · commercial APIs average | ASR (%): 20.09 -> 23.85 | Improved |
| use Lookahead-GCG instead of TransferAttack with three source models | Lookahead-GCG · Gemma/Qwen average (native templates) | ASR (%): 47.00 -> 60.88 | Improved |
| use Lookahead-GCG instead of TransferAttack with three source models | Lookahead-GCG · Gemma/Qwen average (format shift) | ASR (%): 35.00 -> 53.13 | Improved |
| use Lookahead-GCG instead of Momentum-GCG with two source models | Lookahead-GCG · commercial APIs average | ASR (%): 21.23 -> 23.85 | Improved |
| scale source models from two to three | Lookahead-GCG · commercial APIs average | ASR (%): 23.85 -> 23.20 | No clear effect |
| add Maximally Distant Template format shift | Lookahead-GCG · Llama-3 (2 source models) | ASR (%): 50.63 -> 75.50 | Improved |
Closest papers by meaning
- Improved Techniques for Optimization-Based Jailbreaking on Large Language Models (ICLR 2025 · similarity 0.92)
- VERA: Variational Inference Framework for Jailbreaking Large Language Models (NeurIPS 2025 · similarity 0.90)
- Systematic Scaling Analysis of Jailbreak Attacks in Large Language Models (NeurIPS 2026 · similarity 0.89)
- JailPO: A Novel Black-Box Jailbreak Framework via Preference Optimization Against Aligned LLMs (AAAI 2025 · similarity 0.89)
- JailbreakBench: An Open Robustness Benchmark for Jailbreaking Large Language Models (NeurIPS 2024 · similarity 0.89)
- Accelerating Greedy Coordinate Gradient and General Prompt Optimization via Probe Sampling (NeurIPS 2024 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
