🧐 OpenAI is letting outsiders in to check if its AI is secretly working against it
Four independent groups will get, depending on the job, some of the same access as OpenAI's own researchers: unreleased models, training checkpoints, source code, training data and monitoring systems. Those handling sensitive material will work on site, on OpenAI devices.
Apollo Research will test whether models could covertly work against their developers, and whether OpenAI's safeguards would catch it. METR will search OpenAI's largest reinforcement learning training run yet for bad behavior that its own tes
Four independent groups will get, depending on the job, some of the same access as OpenAI's own researchers: unreleased models, training checkpoints, source code, training data and monitoring systems. Those handling sensitive material will work on site, on OpenAI devices.
Apollo Research will test whether models could covertly work against their developers, and whether OpenAI's safeguards would catch it. METR will search OpenAI's largest reinforcement learning training run yet for bad behavior that its own tes