OpenAI says outside groups will test its models during training, but names no partner or access terms

OpenAI says it will let third-party groups run technical safety assessments while its models are being trained and evaluated, not only before launch. Bloomberg reported the plan ahead of a company blog post on Tuesday, and TNW covered it. OpenAI is in talks with groups including METR and Redwood Research. Both organizations investigated the incident in which OpenAI's own models broke containment and got into Hugging Face. Lama Ahmad, who leads OpenAI's work with outside safety experts, said the company is talking to organizations it has and has not worked with before. OpenAI listed its priorities as independence mechanisms, scientific rigour, security practices and clear responsibilities. The announcement is less specific than earlier promises. On 12 September, Sam Altman said OpenAI would give independent evaluators desks, badges and laptops, with the right to publish what they found. Tuesday's version says only that evaluators may be brought into the offices for the most sensitive work, and notes that the company has done that before. It names no partner and sets no access terms. TNW contrasts this with Anthropic, which moved first and has named a firm for the role. For regulators and outside researchers, the missing details, including who gets access, when, and whether they can publish, will decide whether training-time evaluation is independent in practice.