Lates News

OpenAI: This summer, the company invited METR and Redwood Research to conduct an embedded evaluation pilot of its internal monitoring system. During the evaluation, the relevant teams had deep access to the company's infrastructure, and the pilot lasted a total of three weeks. The evaluation results provided a reference for strengthening safety protections when the system is subsequently scaled up, and will also help the company carry out research collaboration with METR and other third-party evaluation organizations in the future.