In short: Apollo Research, METR, Transluce and Goodfire get deep access to OpenAI models, code and data in its first "embedded assessments". Published 21 Nov 2026 by the AI Decoded desk; news dated 16 Nov 2026.

OpenAI has signed agreements with four outside organisations to examine its AI models from the inside. It calls the programme "embedded assessments".

Who's involved

Announced on 16 November 2026, the four are Apollo Research, METR, Transluce and Goodfire. Depending on the job, they get some of the same internal resources OpenAI's researchers use: model checkpoints, source code, training and evaluation data, monitoring systems, and time with the engineers.

What each will do

The catch

Access isn't unlimited. Transluce says it's not comparable to employee-level access, and Apollo says OpenAI may withhold some information. Apollo and Goodfire are paid for their work; METR and Transluce are not.

Why it matters

Independent checks give the public and businesses a clearer view of how labs manage risk, rather than taking the companies' word for it.

FAQs

What is an embedded assessment?

An arrangement where independent experts get deep, extended access to a lab's internal models and systems to check alignment and safeguards.

Will the results be public?

Apollo Research will publish two public reports, and Goodfire says it will publish findings. Not every assessment has a public report deadline.

Sources

Follow AI Decoded on Substack

Weekly plain-English AI news and verified UK events.