Artificial intelligence / news brief
OpenAI gives four outside teams deeper access for embedded alignment assessments
- Event
- Published

OpenAI signed agreements with Apollo Research, METR, Transluce and Goodfire for embedded assessments of model alignment and safeguards. Access can include checkpoints, source code, training data and monitoring systems. METR says its initial pilot is limited to OpenAI's largest reinforcement-learning run and excludes previously disclosed incidents.
Why it matters
The program moves third-party testing inside model development, but its value will depend on assessor independence, scope and publication of findings.
Sources
- Our first phase of embedded assessments (external site)OpenAI · primary source ·
- METR agreement for a limited alignment assessment of OpenAI's largest RL run (external site)METR · coverage ·