Jailbreaking LLMs via Calibration
Fuente:
arXiv
Enregistré dans:
| Auteurs principaux: | , , |
|---|---|
| Format: | Preprint |
| Publié: |
2026
|
| Sujets: | |
| Accès en ligne: | |
| Tags: |
Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
|
| _version_ | 1866911413053685760 |
|---|---|
| author | Lu, Yuxuan Guo, Yongkang Kong, Yuqing |
| author_facet | Lu, Yuxuan Guo, Yongkang Kong, Yuqing |
| contents | Safety alignment in Large Language Models (LLMs) often creates a systematic discrepancy between a model's aligned output and the underlying pre-aligned data distribution. We propose a framework in which the effect of safety alignment on next-token prediction is modeled as a systematic distortion of a pre-alignment distribution. We cast Weak-to-Strong Jailbreaking as a forecast aggregation problem and derive an optimal aggregation strategy characterized by a Gradient Shift in the loss-induced dual space. We show that logit-arithmetic jailbreaking methods are a special case of this framework under cross-entropy loss, and derive a broader family of aggregation rules corresponding to other proper losses. We also propose a new hybrid aggregation rule. Evaluations across red-teaming benchmarks and math utility tasks using frontier models demonstrate that our approach achieves superior Attack Success Rates and lower "Jailbreak Tax" compared with existing methods, especially on the safety-hardened gpt-oss-120b. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2602_00619 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Jailbreaking LLMs via Calibration Lu, Yuxuan Guo, Yongkang Kong, Yuqing Computation and Language Artificial Intelligence Cryptography and Security Machine Learning Safety alignment in Large Language Models (LLMs) often creates a systematic discrepancy between a model's aligned output and the underlying pre-aligned data distribution. We propose a framework in which the effect of safety alignment on next-token prediction is modeled as a systematic distortion of a pre-alignment distribution. We cast Weak-to-Strong Jailbreaking as a forecast aggregation problem and derive an optimal aggregation strategy characterized by a Gradient Shift in the loss-induced dual space. We show that logit-arithmetic jailbreaking methods are a special case of this framework under cross-entropy loss, and derive a broader family of aggregation rules corresponding to other proper losses. We also propose a new hybrid aggregation rule. Evaluations across red-teaming benchmarks and math utility tasks using frontier models demonstrate that our approach achieves superior Attack Success Rates and lower "Jailbreak Tax" compared with existing methods, especially on the safety-hardened gpt-oss-120b. |
| title | Jailbreaking LLMs via Calibration |
| topic | Computation and Language Artificial Intelligence Cryptography and Security Machine Learning |
| url | https://arxiv.org/abs/2602.00619 |