[OpenAI Expands Involvement of Third-Party Organizations in Early Safety Assessments of AI Models] According to Bloomberg, OpenAI has announced that it will allow more third-party organizations to participate in the early stages of AI model training, evaluation, and deployment to conduct technical safety assessments. Previously, the company typically involved external teams only before the release of models to evaluate capabilities and risks. OpenAI stated that such assessments must be independent, scientifically rigorous, equipped with strong safety mechanisms, and have clear divisions of responsibility. OpenAI is currently in discussions with several new and existing partners, including METR and Redwood Research. Some evaluators may be arranged to work on-site at the company in the future to participate in testing the most sensitive aspects.