aiqbench analysis
This research introduces a diagnostic toolkit to identify three specific failure modes in Chain-of-Thought reasoning—post-hoc rationalization, encoded reasoning, and internalized reasoning—that undermine model interpretability. By providing simple metrics and trained "model organisms" for these pathologies, the study offers critical methods for monitoring LLM safety during training without heavy computational overhead.
Diagnosing Pathological Chain-of-Thought in Reasoning Models
arXiv:2602.13904v2 Announce Type: replace Abstract: Chain-of-thought (CoT) reasoning is fundamental to modern LLM architectures and represents a critical intervention point for AI safety. However, CoT reasoning may exhibit failure modes that we note as pathologies, which prevent it from being useful for monitoring. Prior work has identified three distinct pathologies: post-hoc rationalization, where models generate plausible explanations backwards from predetermined answers; encoded reasoning, where intermediate steps conceal information within seemingly interpretable text; and internalized reasoning, where models replace explicit reasoning with meaningless filler tokens while computing internally. To better understand and discriminate between these pathologies, we create a set of concrete metrics that are simple to implement, computationally inexpensive, and task-agnostic. To validate our approach, we develop model organisms deliberately trained to exhibit specific CoT pathologies. Our work provides a practical toolkit for assessing CoT pathologies, with direct implications for training-time monitoring.
Abstract:Chain-of-thought (CoT) reasoning is fundamental to modern LLM architectures and represents a critical intervention point for AI safety. However, CoT reasoning may exhibit failure modes that we note as pathologies, which prevent it from being useful for monitoring. Prior work has identified three distinct pathologies: post-hoc rationalization, where models generate plausible explanations backwards from predetermined answers; encoded reasoning, where intermediate steps conceal information within seemingly interpretable text; and internalized reasoning, where models replace explicit reasoning with meaningless filler tokens while computing internally. To better understand and discriminate between these pathologies, we create a set of concrete metrics that are simple to implement, computationally inexpensive, and task-agnostic. To validate our approach, we develop model organisms deliberately trained to exhibit specific CoT pathologies. Our work provides a practical toolkit for assessing CoT pathologies, with direct implications for training-time monitoring.
Artificial Intelligence (cs.AI)
arXiv:2602.13904 [cs.AI]
(or arXiv:2602.13904v2 [cs.AI] for this version)
https://doi.org/10.48550/arXiv.2602.13904 Focus to learn more arXiv-issued DOI via DataCite
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.
Full text under the publisher's open licence · 229 words
Related coverage
Introducing NV-Reason-CT Open 3D CT VLM for Radiologist Chain-of-Thought Reasoning
NVIDIA · Sep 23, 2026
NVIDIA’s NV-Reason-CT brings AI reasoning to 3D CT scans
Quantum Zeitgeist · Sep 24, 2026
Is Your Model Thinking or Just Stagnating? PUMA: Diagnosing Reasoning Pathology via Phase-Momentum Alignment
arXiv · Jul 21, 2026
MedPAIR: Measuring Whether Physicians and AI Agree on What Matters in Medical QA
Snorkel AI Blog · Oct 2, 2026