According to CryptoSlate, the agreement for Apollo Research and OpenAI's four-month safety evaluation pilot allows Apollo to publicly disclose restrictions or delays in evidence access and to explain the impact of confidentiality redactions on its conclusions. OpenAI may redact confidential supporting materials from public reports, but evaluation conclusions for individual claims cannot be redacted, nor can redactions be requested solely due to unfavorable findings. The evaluations remain limited to agreed-upon models, topics, and timeframes, and certain materials may be withheld for safety, technical, or intellectual property reasons. These are supplemental rules for access and publication and do not signify that the evaluation is complete or that the model has received safety certification.