一项新的研究论文揭示，对大型语言模型（LLMs）进行安全分类的微调可能会无意中产生新的漏洞。虽然这些模型在标准评估中可能表现良好，但它们可能会容易受到规避攻击，这些攻击会保持模型的行为但改变输入。该研究强调了微调如何专门化继承的模型结构，从而导致脆弱的指示器规则，这些规则在保持对保留数据的准确性的同时，也扩大了攻击面。 AI
影响 LLMs的安全微调可能需要更鲁棒的评估方法，这些方法需要考虑语义漂移和保持转换的攻击。
排序理由 该集群包含一篇详细介绍LLM漏洞新发现的研究论文。
AI 生成摘要 · Google Gemini · 来自 2 个来源。 我们如何撰写摘要 →
arXiv:2606.27091v1 Announce Type: cross Abstract: LLMs fine-tuned for security classification are usually evaluated on held-out examples from the same distribution as their training data. We show that this can miss vulnerabilities introduced by fine-tuning itself: models can lear…
LLMs fine-tuned for security classification are usually evaluated on held-out examples from the same distribution as their training data. We show that this can miss vulnerabilities introduced by fine-tuning itself: models can learn token-level indicator semantics that preserve ca…