OpenAI reports evaluate the security benefit boundaries of inference monitoring

OpenAI released the report "Evaluating the Monitorability of the Chain of Thought": Exploring the auditability of model reasoning

OpenAI released a research report "Evaluating Chain-of-Thought Monitorability", which systematically evaluates the monitorability and security impact ...