Citing Apollo Research, TechCrunch reports that OpenAI plans to embed six Apollo safety evaluators internally to investigate whether models attempt to subvert the alignment and control mechanisms of themselves or successor models. The start date for the pilot has yet to be announced, and information access will be restricted to agreed-upon models, timeframes, and evaluation claims, with personnel contact and public evidence also subject to limitations. This arrangement does not constitute a completed evaluation, nor does it imply that the evaluators have been granted unrestricted internal access.