跳到正文
arXiv cs.AI· Qi Tan, Tianyu Cui, Qi Li, Ke Xu·· 4 小时前AI 评分53

安全信息凝聚律论文提出安全泛化误差随模型规模呈U形

Condensing Law of Security Information

AI 导读

Qi Tan 等人的 arXiv 论文(arXiv:2611.17918)提出安全信息凝聚律,论证安全场景下模型越大越好的推论失效,因为 LLM 会隐式修复异常输入再交给安全头。论文用认知能力 K(W) 与安全信息保留率 R(W) 推导三条信息论界,并通过真实 LLM 实验验证安全泛化误差呈U形、在有限最优规模 N* 处最小。

正文 · 原文

View PDF HTML (experimental)

Abstract:Scaling laws indicate that the generalization error of a large language model (LLM) decreases monotonically with model size once data is sufficient, and the corollary that larger models outperform smaller ones is now widely accepted. We argue that this corollary fails in security: decisions there hinge on anomalous inputs, which are long-tailed (e.g., zero-day attacks) and therefore poorly served by likelihood fitting. Worse still, the capability brought about by scale (e.g., inferring the intended whole from a corrupted part) leads a LLM silently \emph{repair} the anomaly before any downstream security head ever sees it. We formalize this as a tradeoff between two quantities: the cognitive capability $K(W)=I(W;P)$ that a model holds about the world distribution, and the security-information retention $R(W)=\frac{I(Z;S)}{I(A;S)}$ that survives in its intermediate representation. We derive three information-theoretic bounds to connect them: an upper bound on restoration accuracy in terms of $K(W)$, an upper bound on $R(W)$ in terms of the restoration error, and a Fano lower bound on security-task error in terms of $R(W)$. These theorems approximately imply that usable security information is a product of how much security information survives and how well it can be extracted, indicating that optimal usable security information can be achieved at a certain model size. We state this as the \emph{condensing law of security information}: usable security information is unimodal in model size, so security generalization error is U-shaped and is minimized at a finite $N^{\star}$ rather than decreasing without bound. The experiments on various real-world LLMs validate the condensing law, indicating that there is an optimal model size to solve security issues.
Subjects: Cryptography and Security (cs.CR); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
Cite as: arXiv:2611.17918 [cs.CR]
  (or arXiv:2611.17918v1 [cs.CR] for this version)
  https://doi.org/10.48550/arXiv.2611.17918

arXiv-issued DOI via DataCite (pending registration)

Submission history

From: Qi Tan [view email]
[v1] Fri, 13 Nov 2026 08:57:58 UTC (169 KB)

来源:arXiv cs.AI · arxiv.org