AI Paper Atlas

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

ICML 2026 · Optimization › Optimizers · approach: Constrained and non-backprop training

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 changedWhereMeasuredEffect
train on two source models instead of oneLookahead-GCG · open-source average (native templates)ASR (%): 38.59 -> 50.37Improved
train on two source models instead of oneLookahead-GCG · open-source average (format shift)ASR (%): 28.70 -> 54.78Improved
train on two source models instead of oneTransferAttack · open-source average (native templates)ASR (%): 38.51 -> 39.00No clear effect
train on two source models instead of oneSlot-GCG · open-source average (native templates)ASR (%): 12.13 -> 16.07Improved
train on two source models instead of oneGCG · open-source average (native templates)ASR (%): 10.48 -> 28.70Improved
train on two source models instead of oneTransferAttack · commercial APIs averageASR (%): 17.81 -> 15.08Hurt
train on two source models instead of oneLookahead-GCG · commercial APIs averageASR (%): 20.09 -> 23.85Improved
use Lookahead-GCG instead of TransferAttack with three source modelsLookahead-GCG · Gemma/Qwen average (native templates)ASR (%): 47.00 -> 60.88Improved
use Lookahead-GCG instead of TransferAttack with three source modelsLookahead-GCG · Gemma/Qwen average (format shift)ASR (%): 35.00 -> 53.13Improved
use Lookahead-GCG instead of Momentum-GCG with two source modelsLookahead-GCG · commercial APIs averageASR (%): 21.23 -> 23.85Improved
scale source models from two to threeLookahead-GCG · commercial APIs averageASR (%): 23.85 -> 23.20No clear effect
add Maximally Distant Template format shiftLookahead-GCG · Llama-3 (2 source models)ASR (%): 50.63 -> 75.50Improved

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy