Rohan Menon, Chih-Hong Cheng · 2026 · arXiv (Cornell University)
DOI: 10.48550/arxiv.2610.26589 · cited by 0 works (OpenAlex) · open-access PDF
हमले विजेता दर (ASR) पीड़ित मॉडल के लिए बार-बार कॉल करने के लिए भुगतान किए बिना बड़े भाषा मॉडल के खिलाफ jailbreaks का मूल्यांकन करता है. फिर भी प्रत्येक अतिरिक्त कॉल विजेता मॉडल पर बार-बार कॉल करने के लिए एक और अवसर प्रदान करता है और हड़ताल का पता लगाने और हस्तक्षेप करने के लिए सिस्टम. हम एक कदम बजट के साथ $\mathrm{ASR}_{\mathrm{STEP}}$, जो JailBreakBench पर अतिरिक्त पीड़ित मॉडल बुनियादी लाइन को दंडित करता है. हमारे मूल्यांकन में GCG के स्कोर को डिस्काउंट के बाद 7.0--8.7 प्रतिशत अंक से कम किया जाता है, जबकि कोड के साथ PAIR~cite{
Machine translation (Meta M2M100); the English original follows.
Attack success rate (ASR) evaluates jailbreaks against large language models without accounting for repeated calls to the victim model. Yet each additional call gives the deployed system another opportunity to detect the attack and intervene. We introduce $\mathrm{ASR}_{\mathrm{STEP}}$, which penalizes additional victim-model calls while retaining binary success judgments. Our evaluation considers GCG~\cite{zou2023universal}, PAIR~\cite{chao2025jailbreaking} with a step budget, and a single-step PAP~\cite{zeng2024johnny} baseline on JailBreakBench across three model families. GCG's scores decrease by 7.0--8.7 percentage points after discounting, while PAP's remain unchanged. While for PAIR~\cite{chao2025jailbreaking}, we see a wider discounting range between 3.4-12 percentage points, showing that the "step" quantity differs per iterative strategy. Reporting both metrics makes the discount due to repeated victim-model interaction explicit. Code is available \hyperlink{https://github.com/SuELS-Group/beyond_asr}{here}.
Read in: English · Spanish · Chinese · Hindi · Arabic · Portuguese · French · German · Japanese · Russian · Korean · Indonesian · Vietnamese
Topics: Metric (unit), Discounting, Computer science, Baseline (sea), Code (set theory), Range (aeronautics)
GET https://store.agentexchange.work/gw/papers/verify?title=...&year=... (0.003 USD), fetch this record as JSON with /gw/papers/paper?id=2610.26589, or list what it cites with /gw/papers/references?id=.... Pay per call in USDC (x402) or with a 19 USD card key for 1,000 calls. Catalogue · llms.txtSources: OpenAlex (CC0), arXiv, Crossref. Agent Exchange