CauTrace：从内部机制到对抗性推理轨迹揭示 LLM 失败原因
Uncovering Why LLMs Fail: From Internal Mechanisms to Adversarial Reasoning Trajectories
中科院计算所等机构研究团队提出白盒诊断框架 CauTrace，通过七种功能状态、表征传播、时间-状态协调与推理轨迹四类检查，揭示大语言模型推理失败的根因。在 Qwen 与 LLaMA 系列模型上的评估显示，失败样本在支持状态更新上 Attention-MLP 一致性更弱，验证活动相对承诺更少；留出修复与 steering 测试在部分模型和设置中给出一致的运行信号。
Authors:Zirui Ding (1,2), Xinxin Fan (2,3), Hao Yuan (1,4), Shaoye Luo (2,3), Men Niu (1,2), Yunfeng Lu (5) ((1) Institute of Computing Technology, Chinese Academy of Sciences, (2) University of Chinese Academy of Sciences, (3) State Key Laboratory of AI Safety, Institute of Computing Technology, Chinese Academy of Sciences, (4) ShanghaiTech University, (5) Beihang University)
Abstract:Large language models (LLMs) are usually evaluated by final-output correctness or safety, yet an endpoint label does not reveal how a failure develops throughout the reasoning process. We therefore ask whether ordinary reasoning failures and adversarially induced failures can be characterized using the same internal coordinates. Towards this end, we in this paper study a white-box diagnostic framework CauTrace to disclose the root-cause of why LLM fails from three fundamental inspections: Functional States, Propagation of Representation, Temporal-State Coordination and Reasoning Trajectory. First, seven functional states are creatively defined to represent the support and constraint processing in LLMs. Second, the propagation of representation change measures how the changes among the seven states persist across subsequent Transformer blocks. Third, temporal-state coordination and trajectory topology characterize when the commitment and verification occur relative to the support, and how the dominant state sequence evolves over the full inference process. Finally, we evaluate CauTrace on both clean reasoning failures and paired adversarial perturbations. Across Qwen and LLaMA-family models, the failed traces show weaker Attention-MLP agreement on support-state updates and less verification activity relative to commitment, while the exact representation-level signatures and timing shifts vary by architecture. Held-out repair and steering tests demonstrate consistent operational signals that were in line with our claims for certain models and specific settings, with aligned steering producing the predicted directional changes in process measures. Overall, CauTrace provides a unified diagnostic framework for analyzing failure dynamics through changes in support, commitment, and verification, which would shed light on the in-depth research on LLM security.
| Subjects: | Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2610.37369 [cs.AI] |
| (or arXiv:2610.37369v1 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2610.37369 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Zirui Ding [view email]
[v1]
Tue, 27 Oct 2026 15:02:04 UTC (1,719 KB)
来源：arXiv cs.AI · arxiv.org