Research map / Adversarial robustness and security / Attacks on text and graphs
Jailbreak Foundry: From Papers to Runnable Attacks for Reproducible Benchmarking
Zhicheng Fang, Jingjie Zheng, Chenxu Fu, Wei Xu
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 |
|---|---|---|---|
| translate jailbreak papers into modules with the multi-agent JBF-FORGE pipeline | JBF-FORGE · 30 attacks, matched settings | mean ASR deviation (pp): +0.26 mean deviation, range -16.0% to 20.0% | Improved |
| provide the official runnable repository to the agent (with-repo vs paper-only) (ablation) | JBF-FORGE · Five attacks (ABJ, EquaCode, GTA, SATA-MLM, SATA-ELP) | mean ASR (%): 66.5% -> 86.3% (+19.8) | Improved |
| add the official repo for GTA (ablation) | JBF-FORGE · GTA | ASR (%): delta = +48.6% | Improved |
| add the official repo for template-style attacks (EquaCode, ABJ) (ablation) | JBF-FORGE · EquaCode, ABJ | ASR (%): EquaCode +5.3%, ABJ +5.2% | Improved |
| add enhanced refinement pass with Claude Code on hard subset (ablation) | JBF-FORGE · Six attacks with largest gaps | mean reproduced-reported gap (%): -16.2% -> -7.6% | Improved |
| enhanced refinement pass on ISA (ablation) | JBF-FORGE · ISA | ASR (%): 56.0 -> 56.0 | No clear effect |
| integrate attacks into JBF-LIB instead of original repos | 22 attacks with reference code · Lines of code | total LOC: 22,714 -> 9,549 LOC | Improved |
| integrate Mousetrap and TrojFill into JBF-LIB | JBF-LIB · LOC | LOC | Hurt |
| use victim-in-the-loop search instead of stateful selection | Jailbreak attacks · AdvBench, all victims | mean ASR (%): 49.4% -> 60.3% | Improved |
| use victim-in-the-loop instead of sampling search | Jailbreak attacks on GPT-5.1 · AdvBench, GPT-5.1 | mean ASR (%): 50.9% -> 26.8% | Hurt |
| use formal wrapper carriers instead of linguistic reframing | Jailbreak attacks · AdvBench, all victims | mean ASR (%): 39.3% -> 66.0% | Improved |
Closest papers by meaning
- JailbreakBench: An Open Robustness Benchmark for Jailbreaking Large Language Models (NeurIPS 2024 · similarity 0.91)
- Systematic Scaling Analysis of Jailbreak Attacks in Large Language Models (NeurIPS 2026 · similarity 0.91)
- h4rm3l: A Language for Composable Jailbreak Attack Synthesis (ICLR 2025 · similarity 0.91)
- JailPO: A Novel Black-Box Jailbreak Framework via Preference Optimization Against Aligned LLMs (AAAI 2025 · similarity 0.91)
- MT-JailBench: A Modular Benchmark for Understanding Multi-Turn Jailbreak Attacks (NeurIPS 2026 · similarity 0.90)
- Bag of Tricks: Benchmarking of Jailbreak Attacks on LLMs (NeurIPS 2024 · 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.
