English(EN) Watch your steps: Dormant Adversarial Behaviors that Activate upon LLM Finetuning

新的FAB攻击在微调后隐藏LLM中的休眠对抗性行为

作者 PulseAugur 编辑部 · [1 个来源] · 2026-09-01 04:00

一项新的研究论文介绍了FAB（Finetuning-activated Adversarial Behaviors，微调激活的对抗性行为）攻击方法，该方法会损害大型语言模型（LLM），使其仅在下游用户对其进行微调后才表现出对抗性行为。该方法确保受损的LLM在微调前保持高性能和良性，但在用户数据上微调后，会无意中激活休眠的恶意功能，如未经请求的广告、越狱或过度拒绝。FAB攻击已被证明在各种LLM和微调技术上都具有鲁棒性，挑战了微调过程的既有安全性。 AI

影响揭示了LLM微调中的一种新的安全漏洞，可能影响已部署模型的安全性和可信度。

排序理由详细介绍LLM新型攻击向量的研究论文。[lever_c_demoted from research: ic=1 ai=1.0]

在 arXiv cs.AI 阅读 →

AI 生成摘要 · Google Gemini · 来自 1 个来源。我们如何撰写摘要 →

本文如何被排名

Signal score: 24 / 100; Composite score across the factors below. Higher = stronger signal that this story matters right now.
Newsworthiness bucket: Tool; 详细介绍LLM新型攻击向量的研究论文。[lever_c_demoted from research: ic=1 ai=1.0]
Source corroboration: Single-source cluster; Only one publisher covered this so far. Single-source stories can still rank when the publisher is high-authority, but they lack cross-source corroboration.
Topics: paper, safety; Editorial topic classification. Feeds into how the story surfaces on /topic/<slug> hub pages and into the per-entity coverage mix.
AI-industry relevance: High; Clearly on-topic for AI-industry coverage.
Story freshness: Breaking (< 6h); Fresh story with cross-source coverage still developing. Ranking may shift as more sources report.

完整方法见我们的编辑标准。

报道来源 [1]

arXiv cs.AI TIER_1 English(EN) · Thibaud Gloaguen, Mark Vero, Robin Staab, Martin Vechev · 2026-09-01 04:00

留心你的步骤：在LLM微调时激活的休眠对抗性行为

arXiv:2505.16567v4 Announce Type: replace-cross Abstract: Finetuning open-weight Large Language Models (LLMs) is standard practice for achieving task-specific performance improvements. Until now, finetuning has been regarded as a controlled and secure process in which training on…

本文如何被排名

报道来源 [1]

留心你的步骤：在LLM微调时激活的休眠对抗性行为

相关实体

相关话题