On September 23, OpenAI, in response to public concerns about the potential hazards of technology, plans to introduce third-party organizations to conduct safety risk reviews of its artificial intelligence models during the early stages of development. The company announced in a blog post on Tuesday that it will allow external organizations to conduct technical safety assessments during the training, evaluation, and launch of new AI models. OpenAI emphasized key elements for ensuring the effectiveness of reviews, including robust independence mechanisms, scientific rigor, sound safety measures, and clear delineation of responsibilities. Lama Ahmad, responsible for safety review collaborations, stated that previously, OpenAI mostly invited external groups to evaluate models before their release, but now, with increasing risks, the company hopes to also prioritize high-risk stages such as training and evaluation.