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.
The paper introduces an unsupervised post‑training objective that maximizes mutual information between chain‑of‑thought (CoT) explanations and model outputs to enhance CoT monitorability. Experiments on three model organisms show that this method recovers most or all lost monitorability in degraded CoT settings and further improves monitorability, unsafe behavior prediction, and consequential reasoning in undegraded policies. The results argue that monitorability should be treated as a trainable attribute and that a dedicated post‑training stage can protect CoT from direct optimization pressure.
OwlSight is a new chain‑of‑thought (CoT) monitoring system that uses a customizable Factor strategy to adapt to a wide range of tasks and monitoring needs. By incorporating fine‑grained and coarse‑grained factors, OwlSight can detect specific behaviors such as email leakage or broader privacy concerns across multiple dimensions like capability, safety, and trustworthiness. Experiments on 15,971 reasoning trajectories show that OwlSight outperforms general‑purpose LLMs by 1.32‑30.57 points at fine granularity and 11.71‑27.81 points at coarse granularity, achieving state‑of‑the‑art results on OpenAI’s CoT monitorability benchmark and demonstrating robustness to model changes, factor rephrasings, CoT compression, and obfuscation.
arXiv:2603. 28590v3 Announce Type: replace Abstract: Large language models (LLMs) can generate chains of thought (CoTs) that are not always causally responsible for their final outputs.
arXiv:2607. 07229v1 Announce Type: new Abstract: Prior work has shown that chain-of-thought (CoT) reasoning is often unfaithful: a model's stated reasoning does not reliably reflect the process that produced its output.
arXiv:2608. 02786v1 Announce Type: new Abstract: AI systems can fail silently.
OpenAI introduces a new framework and evaluation suite for chain-of-thought monitorability, covering 13 evaluations across 24 environments. Our findings show that monitoring a model’s internal reasoning is far more effective than monitoring outputs alone, offering a promising path toward scalable control as AI systems grow more capable.
The paper investigates the monitorability of chain-of-thought (CoT) reasoning in large language models, showing that non-zero mutual information between CoT and output is necessary but not sufficient for effective monitoring. It identifies two approximation errors—information gap and elicitation error—that can degrade monitor performance, and demonstrates that CoT obfuscation can occur even without optimization pressure. The authors propose a label‑free mutual‑information‑based training objective that mitigates CoT obfuscation and reward hacking across multiple environments, outperforming an oracle‑based method.
arXiv:2608. 07528v1 Announce Type: new Abstract: Linear probes detect corrupted context in language models with near-perfect accuracy, yet this does not translate into reliable failure prediction.
Agentic systems entering production typically operate as partially integrated assemblies where structural defects, not task-level errors, dominate the failure landscape. At this maturity level, task-level error detection may be infeasible: structural failure modes mask the signal that task-level monitors are designed to detect.
AI Morbidity and Mortality (AI M&M) is a structured, blameless framework designed to review clinical AI failures. It combines standardized case intake, evidence preservation, investigator reconstruction, tool‑in‑loop attribution, and corrective‑action tracking, classifying each event across four linked dimensions: Trigger, Mechanism, Clinical Pathway, and Corrective Action. The authors demonstrate the framework with five outpatient medication and clinical decision‑support cases, achieving full agreement among reviewers on all classification axes.
arXiv:2606. 02494v1 Announce Type: cross Abstract: Agentic systems entering production typically operate as partially integrated assemblies where structural defects, not task-level errors, dominate the failure landscape.