monitorability

Yale researchers establish causal upper and lower bounds on CoT monitorability metrics, showing that the minimal-criterion TPR metric can overestimate intervention-caused behavior detection, and propose three intervention-aware post-training methods (off-policy self-distillation, RL, on-policy self-distillation) that improve LLM reasoning-trace monitorability.

OpenAI 探讨了 g-mean² 之外用于衡量模型可监控性的补充指标，针对结果、过程与干预三类评估，分析任务混合差异与部署时控制使用场景对单一监控性数值的局限。

OpenAI发布声明回应三名离职的安全研究员Jasmine、Mikita和Tomek，称解雇原因是违反敏感信息处理政策而非压制安全发声，并宣布正在敲定第三方安全评估合作协议。

A tweet satirizing AI safety discourse around losing chain-of-thought monitorability due to opaque internal reasoning, riffing on a quote attributed to Miles Brundage.

This paper investigates whether training LLMs for shorter, more efficient chain-of-thought reasoning harms the faithfulness and monitorability of that reasoning. It finds faithfulness generally declines due to reduced model consistency, while monitorability remains relatively robust even with much shorter CoT outputs.

This article argues for preserving Chain of Thought transparency in AI models to maintain monitorability for safety, highlighting examples like OpenAI's GPT-6 Astra and the need for scientific methods to ensure reasoning remains interpretable.

This document operationalizes a measure called 'opaque serial depth' to quantify unverbalized serial cognition in AI models, proposing natural-language-rooted nodes as interpretable bottlenecks to enhance oversight of latent reasoning.

This paper shows that length-penalized reinforcement learning shortens chain-of-thought reasoning but also reduces monitorability, as compressed chains preferentially remove cues that reveal influences behind answers, making it harder to detect steering or biases.