Abstract:Jailbreak prompts can trigger harmful completions on aligned LLMs, In accordance, safety steering has been proposed: test-time activation interventions that steer jailbreak activations to trigger refusal while preserving benign utility. However, existing steering methods are fundamentally supervised and tied to a static, limited training set, whereas real jailbreaks evolve and are often out-of-distributed from the training set, leading to failures on unseen attacks. In this paper, we tackle the failure on unseen jailbreaks problem, base on unsupervised latent direction discovery. We propose a bi-level adversarial training framework for zero-shot jailbreak defense. In the inner step, we simulate diverse jail-broken activations by extrapolating from refusal-state harmful-request activations via unsupervised latent direction discovery, which expands the coverage of real jailbreak activation subspaces. In the outer step, we train a potential-induced steering field to push these adversarial jailbroken states into refusal regions while keeping benign unchanged. Across three LLMs and six classical jailbreak families, our method achieves strong defense with attack success rates mostly below 5%, and rising subspace coverage throughout training helps explain the improved generalization.
| Comments: | accepted by ICML 2026 |
| Subjects: | Cryptography and Security (cs.CR); Machine Learning (cs.LG) |
| Cite as: | arXiv:2605.24535 [cs.CR] |
| (or arXiv:2605.24535v1 [cs.CR] for this version) | |
| https://doi.org/10.48550/arXiv.2605.24535 arXiv-issued DOI via DataCite (pending registration) |
From: Luoyu Chen [view email]
[v1]
Sat, 23 May 2026 12:07:38 UTC (11,626 KB)
此内容由惯性聚合(RSS阅读器)自动聚合整理，仅供阅读参考。 原文来自 — 版权归原作者所有。