Researcher, Alignment CoT Monitorability

ABOUT THE TEAM The CoT Monitorability team at OpenAI studies whether and when the chain-of-thought of frontier reasoning models is monitorable enough to support scalable oversight. We study how to measure monitorability https://openai.com/index/evaluating-chain-of-thought-monitorability/, which training mechanisms affect monitorability, and speculative methods to improve monitorability. While we mostly focus on CoT monitorability at the moment, we care more generally about any form of monitorabi… Full description ↗