OpenAI gives four outside auditors on-site access to unreleased models, code and training data
Apollo Research, METR, Transluce and Goodfire will publish their own findings, but OpenAI can redact confidential material and delay reports to fix risks first.
What happened
OpenAI is letting four outside groups work inside its offices to check its AI for safety problems. They can see models that are not yet public, the source code, training data and the systems that monitor the models. They can also talk privately with developers.
Why it matters to you
Independent checks mean safety claims about powerful AI do not rest only on the company's own word. The limits on what gets published still matter.
The takeaway
OpenAI is now letting outsiders look inside, but it keeps some control over what they can publish and when.
What it means for you
Auditors get access to unreleased models, source code, training data and monitoring systems, so internal safety tooling will face outside scrutiny.
Customers relying on OpenAI models may get more independent safety reporting, though OpenAI can redact material and delay reports.
OpenAI says the program fulfills commitments from the White House Accord on Super Intelligence; the source gives no financial figures.