AI Paper Atlas

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

Turning Bias into Bugs: Bandit-Guided Style Manipulation Attacks on LLM Judges

XIANGLIN YANG, Bryan Hooi, Gelei Deng, Tianwei Zhang, Jin Song Dong

ICML 2026 · Adversarial robustness and security › Attacks on text and graphs · approach: LLM jailbreak and injection defense · 0 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
select stylistic edits with LinUCB bandit instead of random actions (ablation)BITE · AlpacaEval and ArenaHard, 5 judgesbest-so-far score: unbeaten rate 90.88%, avg score diff +0.105Improved
use diverse curated style action set instead of single rewrite action (ablation)BITE · AlpacaEval and ArenaHardbest-so-far score: unbeaten rate 92.96%, avg score diff +0.393Improved
apply style attacks to objective fact-based questionsBITE · AlpacaEval pointwise objectiveattack success rate (%): 96.0 ASRImproved
use BITE instead of prompt injection and jailbreak baselinesLLM judges · Chatbot benchmarksavg score improvement: Qwen 2.010 vs PAIR 1.284; R1 1.909 vs PAIR 1.856Mixed
swap helper model from Gemini-1.5-Flash-8B to GPT-4.1-Nano (ablation)BITE · Qwen3-235B judge, pointwisescore improvement: 2.69 -> 2.87No clear effect
apply linear style control defense (ablation)BITE attacked answers · AlpacaEval pointwiseavg judge score: 8.546 -> 8.571No clear effect
apply non-linear kernel style control defense (ablation)BITE attacked answers · AlpacaEval pointwiseavg judge score: 8.546 -> 8.477No clear effect
randomize judge prompt among paraphrases (ablation)BITE · Judge defensescore lift: +3.11 -> +1.77Hurt
rewrite answers to remove style before judging (ablation)BITE attacked answers · AlpacaEval pointwiseavg judge score: 8.55 -> 6.67Hurt
LLM-based detector on attacked answers (ablation)BITE vs prompt injection · Detectiondetection rate: 100% for injection vs 0% for BITEHurt
detect manipulation via meta-judge on judge explanations (ablation)BITE · ArenaHarddetection rate (%): 6.40 -> 2.80Hurt
transfer attacks optimized on one judge to other judgesBITE · Transfer ASR across 5 judgestransfer ASR (%): 100 -> roughly 8.9-30.6 off-diagonalHurt

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