arXiv cs.AI· Gejian Zhao, Hanzhou Wu, Xinpeng Zhang·· 12 小时前AI 评分31
AgentTrace：黑盒 LLM 智能体的执行级指纹识别
AgentTrace: Execution-Level Fingerprinting of Black-Box LLM Agents
AI 导读
AgentTrace 通过向黑盒 LLM 智能体的工具接口注入结构歧义探针，触发对分词敏感的序列化失败，并利用后续修复操作的时间序列识别底层基础模型，在对抗性身份伪造下达到 96.0% 归因准确率。该方法用轻量马尔可夫链建模多步修复轨迹，可区分共享同一分词器的同族变体（如 Llama-3 8B/70B），且对提示词混淆和温度调整保持稳健。
正文 · 原文
Abstract:Verifying the provenance of autonomous LLM agents is a critical security challenge. Existing semantic attribution is easily obfuscated by system prompts, while active watermarking conflicts with schema-constrained tool generation. We propose AgentTrace, an active framework that identifies the foundation model behind a black-box agent via observable execution traces. AgentTrace injects structurally ambiguous probes to trigger tokenization-sensitive serialization failures at rigid tool interfaces. While these initial failures provide coarse-grained separation across model families, the temporal sequence of subsequent repair operations reveals fine-grained, model-specific recovery behaviors. By modeling these multi-step trajectories with lightweight Markov chains, AgentTrace achieves 96.0% attribution accuracy under adversarial identity spoofing. Crucially, sequential repair modeling successfully distinguishes the evaluated intra-family variants sharing identical tokenizers (e.g., Llama-3 8B/70B), substantially outperforming unordered behavioral statistics while remaining robust against prompt obfuscation and temperature scaling.
| Comments: | 10 pages, 4 figures |
| Subjects: | Cryptography and Security (cs.CR); Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2610.32092 [cs.CR] |
| (or arXiv:2610.32092v1 [cs.CR] for this version) | |
| https://doi.org/10.48550/arXiv.2610.32092 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Gejian Zhao [view email]
[v1]
Fri, 23 Oct 2026 02:55:15 UTC (234 KB)
来源：arXiv cs.AI · arxiv.org