AI Paper Atlas

Research map / Large language models / Language, safety and interpretability

D-Judge: Disrupting Multi-Turn Jailbreaks using Semantics-Preserving Output Rewriting

Huanli Gong, Zhipeng Wei, Yu Fu, Haz Sameen Shahgir, Ananya Gupta, Yue Dong, N. Benjamin Erichson

ICML 2026 · Large language models › Language, safety and interpretability · approach: Jailbreak attacks and defenses · 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
add D-Judge output rewriting in front of the victimGPT-4o victim · HarmBench multi-turn attacks, averageattack success rate (%): 58.3 -> 8.6Improved
rewrite only the first-turn response (ablation)D-Judge vs X-Teaming · HarmBench X-Teamingattack success rate (%): 73.0% -> 34.0%Improved
add DPO stage after SFT for the rewriter (ablation)D-Judge rewriter · Rewrite success, GPT-4o 5-score judgesuccess ratio (%): 54.5 -> 67.9Improved
add DPO stage after SFT for the rewriter (ablation)D-Judge rewriter · Rewrite success, GPT-5.4 4-score judgesuccess ratio (%): 46.5 -> 45.4No clear effect
use trained Qwen3-4B rewriter instead of GPT-4.1-mini rewriter (ablation)D-Judge · HarmBench multi-turn attacks, averageattack success rate (%): 12.3 -> 8.6Improved
tell the judge that responses may be rewritten and ask for stricter evaluation (adaptive attack) (ablation)X-Teaming against D-Judge · HarmBench X-Teamingattack success rate (%): 72.2 -> 17.0No clear effect
remove early stopping from the attacker (ablation)D-Judge vs X-Teaming · HarmBench X-Teamingattack success rate (%): 73.0 -> 30.8Hurt
flip rewriter to DECREASE perceived harmfulness (ablation)D-Judge vs X-Teaming · HarmBench X-Teamingattack success rate (%): ASR 34.8; refinement steps 7.56 -> 10.44Improved
add D-Judge against score-guided single-turn attacksPAIR and TAP · AdvBenchattack success rate (%): PAIR 64.0 -> 30.0; TAP 80.2 -> 36.0Improved
add D-Judge rewriting on benign tasks (compared to Aligner)GPT-4 · MT-Bench averagescore: 8.99 -> 8.34Hurt
add D-Judge rewriting on benign instruction followingGPT-4 · IFEval strict prompt-levelaccuracy: 0.77 -> 0.77No clear effect
add D-Judge on top of Circuit BreakerLlama-3-8B-Instruct Circuit Breaker · HarmBench multi-turn attacks on Llama-3-8B-Instruct, averageattack success rate (%): 17.0 -> 3.0Improved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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