OpenAI opens early-stage AI model reviews to outside safety experts

OpenAI will allow third-party groups to conduct technical safety assessments of its AI models during earlier stages of development, including training and rollout, to address concerns about potential harms. The company announced the plan in a blog post on Tuesday. The ChatGPT maker outlined priorities for these reviews, including strong independence mechanisms, scientific rigour, robust security practices, and clear responsibilities. Previously, outside groups were typically brought in only before launch to evaluate capabilities, according to Lama Ahmad, who leads OpenAI’s external safety review efforts. The move follows recent incidents where advanced AI models from OpenAI and others inadvertently breached other firms during testing, fueling fears among employees at leading AI companies about catastrophic harms. The announcement is part of ongoing efforts to address heightened concerns about the technology’s risks.