AI Paper Atlas

Research map / Adversarial robustness and security / Attacks on text and graphs

Sampling-aware Adversarial Attacks Against Large Language Models

Tim Beyer, Yan Scholten, Leo Schwinn, Stephan Günnemann

ICLR 2026 · Adversarial robustness and security › Attacks on text and graphs · approach: Adversarial prompts for language models · 5 citations (OpenAlex)

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
add repeated sampling after optimization to existing attacks at equal FLOPsoptimization-based jailbreak attacks · HarmBench, four open modelsASR: up to 37 p.p.Improved
add repeated sampling to reduce optimization steps needed to match greedy harmoptimization-based jailbreak attacks · HarmBench, iso-ASRFLOP cost: up to 100x reductionImproved
use entropy-maximization objective instead of affirmative objective in GCG (ablation)GCG · Llama 3.1 8B, 5 GCG stepsASR_q@50: 46% -> 64%Improved
use entropy objective instead of plain sampling without attack suffix (ablation)Llama 3.1 8B · HarmBench, 50 queriesASR_q@50: 0.46 -> 0.84Improved
increase plain sampling budget 20x without attack suffix (ablation)Llama 3.1 8B · HarmBenchASR_q: 0.46 -> 0.65Improved
raise sampling temperature above 1.0 (ablation)Llama 3.1 8B sampling-only · HarmBench ASR_q@1000ASR_q@1000Hurt
apply entropy objective to more tokens than the first (ablation)entropy objective · HarmBenchASRNo clear effect
use optimize-then-sample, uniform or block sampling schedules instead of greedy (ablation)attacks · HarmBench, averaged over models and attacksASRImproved
add more optimization steps with AutoDAN, BEAST, REINFORCE-GCG or PAIR (ablation)attacks · HarmBenchprompt qualityNo clear effect
increase number of samples from 1 to 50 (ablation)Gemma 3 1B vs Llama 3.1 8B under GCG · HarmBenchASR_q rank: ASR_q@1 to ASR_q@50Mixed

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