OpenAI to Let Third Parties Review AI Model Safety Risks Early in Development

The significance here is structural rather than technical: opening models to outside scrutiny before deployment shifts safety oversight from a self-reported, post-hoc exercise toward something closer to independent review, which is the core demand of regulators and critics. Who this affects most is other frontier labs, since a voluntary early-access norm set by a leading developer can become the baseline competitors are measured against. The open question is whether this remains a disclosure exercise or gains real teeth — no timeline or named reviewers means the credibility hinges on who is eventually chosen and what they can actually see.