AI Paper Atlas

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

RLCracker: Evaluating the Worst-Case Vulnerability of LLM Watermarks with Adaptive RL Attacks

Hanbo Huang, Yiran Zhang, Hao Zheng, Xuan Gong, Yihan Li, Lin Liu, Zhuotao Liu, Shiyu Liang

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
add role-assignment system prompt to paraphrase attackQwen3-4B paraphraser · EWD 500 tokensESR (%): 39.3 -> 54.3Improved
add role-assignment system prompt to paraphrase attackQwen3-4B paraphraser · KGW selfhash 500 tokensESR (%): 70.0 -> 79.3Improved
prompt paraphraser to reason before rewriting (Think)Qwen3-4B paraphraser · EWD 500 tokensESR (%): 39.3 -> 76.0Improved
prompt paraphraser to reason before rewriting (Think)Qwen3-4B paraphraser · KGW selfhash 1500 tokensESR (%): 36.0 -> 78.0Improved
add more enforced reasoning steps at test timeQwen3-8B paraphraser · SIR and PF 500 tokensESR (%)Improved
train paraphraser with RLCracker instead of GPT-4o promptingQwen2.5-3B-Instruct · Unigram 1500 tokensESR (%): 6.75 -> 98.5Improved
train with RLCracker instead of SIRAQwen2.5-3B-Instruct · Unigram 1500 tokensP-SP score: 0.47 -> 0.92Improved
apply RLCracker to a Think+SysP baseline on Unigram 500 tokensQwen3-4B · Unigram 500 tokensESR (%): 38.8 -> 73.5Improved
increase training samples from 50 to 100 (ablation)RLCracker Qwen3-0.6B · EWD 250 tokensESR (%): 82.5 -> 87.8Improved
increase training samples from 100 to 200 (ablation)RLCracker Qwen3-0.6B · EWD 500 tokensESR (%): 91.5 -> 92.0No clear effect
increase training token length per sample from 250 to 500 (ablation)RLCracker Qwen3-0.6B · EWD (50 samples)ESR (%): 82.5 -> 86.5Improved
train on mixed-key watermarked data instead of single key (ablation)RLCracker Qwen3-0.6B · EWD 500 tokensESR (%): 91.5 -> 88.8Hurt

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