According to CNBC, OpenAI has signed agreements with Apollo Research, METR, Transluce, and Goodfire to conduct pilot evaluations of model behavior and safeguards, respectively. Goodfire will focus on training monitoring, debugging, and alignment, while Transluce will first assess post-training alignment tests. The evaluators are permitted to publish their conclusions, though OpenAI retains the right to redact confidential information. Apollo and Goodfire receive compensation that is independent of the results, whereas METR and Transluce will not charge evaluation fees. The agreements do not imply that the evaluations have been completed or that the models have received safety certification.