跳到正文
arXiv cs.AI· Zhaojia Wang, Jiucui Lu, Zheng Li·· 4 小时前AI 评分52

arXiv 论文实证研究:LLM 能否通过权重空间编程为神经网络植入后门

Can LLMs Write Backdoors? An Empirical Security Study of Weight-Space Programming

AI 导读

arXiv 论文(arXiv:2611.17395)实证研究 LLM 能否依据自然语言指令直接修改模型权重植入后门,无需任务数据、前向传播或梯度。

正文 · 原文

View PDF HTML (experimental)

Abstract:Recent advances in weight-space programming (WSP) have demonstrated that large language models (LLMs) can directly inspect and modify neural network weights without conventional gradient-based training. While this capability opens new possibilities for model improvement, it also raises an important security question: Can LLMs exploit weight-space programming to inject backdoors into neural networks? To answer this question, we present an empirical security study of LLM-driven weight-space programming for backdoor injection. Specifically, we investigate whether general-purpose LLMs can directly modify the weights of target models according to natural-language instructions, without accessing task-specific data, performing forward passes, or computing gradients. We evaluate five LLMs as weight editors across three target architectures: an MLP on MNIST, ResNet-18 on CIFAR-10, and Qwen2.5-0.5B-Instruct on AG News. Our experiments reveal three key findings. First, LLM-driven weight editing can successfully inject backdoors across different architectures while largely preserving clean performance. Second, successful attacks can require remarkably few parameter modifications, including only six parameters for the MLP and 0.02% of the weights for a fine-tuned Qwen model. Third, editing capabilities vary substantially across LLMs and target architectures: although all five LLMs successfully backdoor the MLP, only GPT-6 Astra succeeds across all three targets. These findings highlight an emerging security risk of weight-space programming and underscore the importance of safeguarding LLM-driven weight editing.
Subjects: Cryptography and Security (cs.CR); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
Cite as: arXiv:2611.17395 [cs.CR]
  (or arXiv:2611.17395v1 [cs.CR] for this version)
  https://doi.org/10.48550/arXiv.2611.17395

arXiv-issued DOI via DataCite (pending registration)

Submission history

From: Zheng Li [view email]
[v1] Thu, 12 Nov 2026 16:41:07 UTC (6,586 KB)

来源:arXiv cs.AI · arxiv.org