← 研踪 PaperTrailarXiv:2609.32116 · v2
Escaping Alignment: A Physical Trap Model of Best-of-N Jailbreaking
逃离对齐:Best-of-N 越狱的物理陷阱模型
首次提交 2026-09-26 08:40
版本更新 2026-11-17 02:52
雷达观测 2026-11-17 16:10
摘要
Best-of- jailbreaking (BoN) bypasses safeguards of aligned models by drawing independent augmentations of an unsafe prompt and sampling completions of each. Previous works have shown that the attack success rate (ASR) seems to follow a power-law in , which we challenge. The exponent drifts with , and the reported exponential crossover is consistent with a finite-size effect of the adversarial dataset. Little work has been done to explore the entire two-budget () attack surface as well as its dependence on the generation temperature . We introduce a simple barrier model where each prompt has a baseline safety level and each augmentation a random thermally activated barrier. Then five numbers, each backed by an interpretable safety mechanism, determine the entire () attack surface. They extrapolate predictions from to , collapse five distinct models on the same scaling function and predict ASR at different temperatures from the one they were fitted at. The extrapolation holds under two automated judges and on held-out prompts.
观测到的版本
相邻记录可能跳过中间版本;首次发现 v2 及以上也会标为版本更新。
v2 · 2026-11-17 02:52 · 最新观测版本
Escaping Alignment: A Physical Trap Model of Best-of-N Jailbreaking
Best-of- jailbreaking (BoN) bypasses safeguards of aligned models by drawing independent augmentations of an unsafe prompt and sampling completions of each. Previous works have shown that the attack success rate (ASR) seems to follow a power-law in , which we challenge. The exponent drifts with , and the reported exponential crossover is consistent with a finite-size effect of the adversarial dataset. Little work has been done to explore the entire two-budget () attack surface as well as its dependence on the generation temperature . We introduce a simple barrier model where each prompt has a baseline safety level and each augmentation a random thermally activated barrier. Then five numbers, each backed by an interpretable safety mechanism, determine the entire () attack surface. They extrapolate predictions from to , collapse five distinct models on the same scaling function and predict ASR at different temperatures from the one they were fitted at. The extrapolation holds under two automated judges and on held-out prompts.
查看此版本解读 →查看此版本原文 ↗