analiza aiqbench
The paper proposes CauTrace, a white-box method that tracks how an LLM's internal representations move through "functional states" across Transformer layers and the reasoning process, rather than judging only the final answer. Its reported signals, such as weaker attention-MLP agreement and less verification relative to commitment in failed traces, are tested on Qwen and LLaMA-family models, and the authors note that the specific signatures vary by architecture. Readers should treat the results as preliminary, model-specific diagnostics, not a general account of why LLMs fail or a validated security tool.
Uncovering Why LLMs Fail: From Internal Mechanisms to Adversarial Reasoning Trajectories
De la arXiv
Text integralarXiv:2610.37369v1 Announce Type: new 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.
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.
Artificial Intelligence (cs.AI)
arXiv:2610.37369 [cs.AI]
(or arXiv:2610.37369v1 [cs.AI] for this version)
https://doi.org/10.48550/arXiv.2610.37369 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 347 cuvinte
Subiecte conexe
Across Viewpoints and Time: Trajectory-Generalizable Physical Attacks on World Models and Video Understanding Models
arXiv · 15 oct. 2026
MADS: Multi-Agent Dialog Simulation for Robust Conversational Agents
arXiv · 12 oct. 2026
Anchoring Adversarial Trajectories to Data Manifolds: A Bilevel Transfer Optimization Framework
arXiv · 1 oct. 2026
Adversarial Attacks on Downstream Weather Forecasting Models: Application to Tropical Cyclone Trajectory Prediction
arXiv · 16 nov. 2026