arXiv cs.AI· Jonathan Pan, Yidi Yuan, Xin Wei Chia, Xinhui Su, Vincent Tan, Brenda Ong·· 4 小时前AI 评分48
从字节到行为:基于在线 GPU 隐状态的 LLM 请求级取证归因
From Bytes to Behavior: Request-Level Attribution from Live GPU Hidden States
AI 导读
一项 arXiv 研究提出用非侵入式调试器从在线 GPU 进程中按层提取残差流激活,无需预先了解模型架构参数或权重,可在不重启、不修改 LLM 服务的情况下取证。该方法在 9 类架构、33 个开源 decoder-only 模型上验证,恢复字节完全精确。训练于标准框架激活上的线性探针应用到恢复状态后,区分拒绝与服从的 AUROC 达 0.88,为审查已部署 AI 系统的运行行为提供新证据来源。
正文 · 原文
Abstract:Existing memory forensics for machine learning systems typically targets static artifacts, such as model weights or prompt logs, rather than the transient computational states of a live inference process. This gap limits an examiner's ability to analyze the specific behavior of a running large language model (LLM) service that cannot be restarted or modified. We present a method for acquiring per-layer residual-stream activations from a live GPU process using a non-modifying debugger, without requiring prior knowledge of the model's architecture parameters or weights. The approach employs a general six-stage process with platform-specific adapters, utilizing either a PyTorch allocator clock or a kernel-launch clock to identify and read tensor allocations from video memory. We validate this acquisition on 33 open-source decoder-only models across nine architecture classes, demonstrating that the recovered bytes are byte-exact and usable for downstream analysis. As a use case, we apply linear probes trained on standard framework activations to the recovered states, successfully distinguishing refusal from compliance with an AUROC of 0.88 on held-out prompts. This work establishes that live, observational access to intermediate hidden states is feasible on named inference platforms, providing a new source of evidence for examining the operational behavior of deployed AI systems.
| Comments: | 7 pages, 1 figure, 4 tables |
| Subjects: | Cryptography and Security (cs.CR); Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2611.18272 [cs.CR] |
| (or arXiv:2611.18272v1 [cs.CR] for this version) | |
| https://doi.org/10.48550/arXiv.2611.18272 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Jonathan Pan [view email]
[v1]
Fri, 13 Nov 2026 15:04:05 UTC (37 KB)
来源:arXiv cs.AI · arxiv.org