← 研踪 PaperTrailACL Anthology · ACL 2026:acl:2026.acl-long.1310 · 记录 #1
OASIS: Mitigating Harmful Fine-tuning Attacks on LLMs via Orthogonal and Adaptive Safety Alignment Strategy
来源日期（仅到月份） 2026-07
来源记录日期 2026-07
雷达观测 2026-11-01 21:54
摘要
The “Fine-Tuning-as-a-Service” paradigm exposes large language models to catastrophic safety degradation from less harmful samples. Alignment-stage defenses address this by proactively injecting adversarial perturbations to bolster the model’s inherent robustness against harmful drift. However, existing methods rely on perturbation directions that often conflict with harmful gradients, inadvertently facilitating the acquisition of malicious features rather than suppressing them. To address this issue, we propose Orthogonal and Adaptive Safety Alignment Strategy (OASIS) to mathematically decouple safety enforcement from harmful feature acquisition. By projecting perturbations orthogonal to harmful gradients and concentrating optimization on adaptively selected safety-critical layers, OASIS effectively resolves directional conflicts while maximizing parameter efficiency. Extensive experiments on four LLMs across three datasets (SST2, GSM8K, and AGNews) demonstrate that OASIS reduces the Harmful Score by approximately 60% compared to competitive baselines, while maintaining stable downstream task utility.
论文发现出处
这些页面提供论文线索，不代表同行评审或质量认证。
观测到的版本
平台记录号代表本站观测到的元数据变化，不是作者修订号。来源日期仅到月份或年份时，按该月或该年首日存档。
记录 #1 · 2026-07 · 最新观测版本
OASIS: Mitigating Harmful Fine-tuning Attacks on LLMs via Orthogonal and Adaptive Safety Alignment Strategy
The “Fine-Tuning-as-a-Service” paradigm exposes large language models to catastrophic safety degradation from less harmful samples. Alignment-stage defenses address this by proactively injecting adversarial perturbations to bolster the model’s inherent robustness against harmful drift. However, existing methods rely on perturbation directions that often conflict with harmful gradients, inadvertently facilitating the acquisition of malicious features rather than suppressing them. To address this issue, we propose Orthogonal and Adaptive Safety Alignment Strategy (OASIS) to mathematically decouple safety enforcement from harmful feature acquisition. By projecting perturbations orthogonal to harmful gradients and concentrating optimization on adaptively selected safety-critical layers, OASIS effectively resolves directional conflicts while maximizing parameter efficiency. Extensive experiments on four LLMs across three datasets (SST2, GSM8K, and AGNews) demonstrate that OASIS reduces the Harmful Score by approximately 60% compared to competitive baselines, while maintaining stable downstream task utility.
查看此版本解读 →查看此版本原文 ↗