OpenAI gives 4 outside safety groups sustained internal access
Apollo Research, METR, Transluce and Goodfire will assess OpenAI’s alignment and safeguards during training and internal use, with access that can include checkpoints, code and training data.
Source published · In the Daily Brief of 18 November 2026
Filed under AI safety, AI industry
OpenAI announced on 16 November that it had signed agreements with four independent organisations for the first phase of what it calls embedded assessments. Their access can include model checkpoints, source code, training data and monitoring systems, sustained over an extended period, according to Unite.AI.s account of the announcement.
Each group has its own mandate. Apollo Research will examine whether models could covertly act against their developers’ intentions and whether OpenAI’s safeguards would catch it, testing four propositions in two stages. Goodfire will embed researchers in interpretability, the analysis of model internals. OpenAI’s Safety and Security Committee receives unredacted reports and can speak privately with the assessors.