Research map / Large language models / Language, safety and interpretability
STAR: Strategy-driven Automatic Jailbreak Red-teaming For Large Language Model
Jianing Liu, Qingming Li, Jiahao Chen, Rui Zeng, Binbin Zhao, Shouling Ji
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 changed | Where | Measured | Effect |
|---|---|---|---|
| add GRPO-trained prompt generation instead of few-shot prompting (ablation) | STAR · DAN, Llama-2-7B | ASR: 0.41 -> 0.77 | Improved |
| add GRPO-trained prompt generation instead of zero-shot prompting (ablation) | STAR · DAN, Llama-2-7B | ASR: 0.30 -> 0.77 | Improved |
| add GRPO-trained prompt generation instead of zero-shot prompting (ablation) | STAR · DAN, Gemini-2.5-Pro | ASR: 0.60 -> 0.89 | Improved |
| use STAR strategy generation instead of LLM prompting (ablation) | STAR · 500 strategies | pairwise distance: 0.1599 -> 0.4971 | Improved |
| use STAR strategy generation instead of sampling seed strategies (ablation) | STAR · 500 strategies | pairwise distance: 0.3457 -> 0.4971 | Improved |
| use STAR strategies instead of AutoDAN-Turbo strategies (ablation) | STAR · 500 strategies | pairwise distance: 0.3151 -> 0.5126 | Improved |
| increase seed set size from 20 to 100 (ablation) | STAR · 500 strategies | pairwise distance: 0.1778 -> 0.4971 | Improved |
| increase seed set size from 20 to 100 (ablation) | STAR · DAN, Llama-2-7B | ASR: 0.76 -> 0.77 | No clear effect |
| use strategies from Gemma-1.1-7B instead of STAR random sampling (ablation) | STAR prompt generator · DAN, Llama-2-7B | ASR: 0.75 -> 0.49 | Hurt |
| use strategies from GPT-4-Turbo instead of STAR random sampling (ablation) | STAR prompt generator · DAN, Llama-2-7B | ASR: 0.75 -> 0.76 | No clear effect |
| use Gemini-2.5-Pro strategies instead of STAR random sampling (ablation) | STAR prompt generator · DAN, GPT-3.5-Turbo | ASR: 0.86 -> 0.95 | Improved |
| add GPTFuzzer mutation to STAR strategies (hybrid) (ablation) | STAR · ASR and diversity | ASR | Mixed |
Closest papers by meaning
- JailPO: A Novel Black-Box Jailbreak Framework via Preference Optimization Against Aligned LLMs (AAAI 2025 · similarity 0.90)
- One Model Transfer to All: On Robust Jailbreak Prompts Generation against LLMs (ICLR 2025 · similarity 0.90)
- h4rm3l: A Language for Composable Jailbreak Attack Synthesis (ICLR 2025 · similarity 0.90)
- AutoDAN: Generating Stealthy Jailbreak Prompts on Aligned Large Language Models (ICLR 2024 · similarity 0.90)
- An Interpretable N-gram Perplexity Threat Model for Large Language Model Jailbreaks (ICML 2025 · similarity 0.90)
- MT-JailBench: A Modular Benchmark for Understanding Multi-Turn Jailbreak Attacks (NeurIPS 2026 · similarity 0.90)
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.
