OpenAI has signed agreements with Apollo Research, METR, Transluce and Goodfire for what it calls the first phase of embedded assessments, a program that places independent safety researchers inside its model-development process. OpenAI announced the agreements on November 16, describing the program as giving outside experts deep access to its internal models and systems to investigate alignment and test its safeguards.
Access may include model checkpoints, source code, training and evaluation data, and monitoring systems, along with time with the researchers and engineers responsible. OpenAI keeps the right to redact confidential information, and said it would consider alternatives, such as pairing assessors with its own researchers, where direct access could expose its most sensitive intellectual property, such as model weights or architecture.
Each organization runs a separate pilot. Apollo Research will assess whether OpenAI's models could covertly work against developers' intentions and whether safeguards would catch it. Its assessment tests four claims: that training did not incentivize subversion, that strong evaluations and red-teaming do not trigger it, that models have not attempted it during internal use, and that reasoning linked to such behavior would be detectable.
Key facts
- 01OpenAI announced embedded assessments with Apollo Research, METR, Transluce and Goodfire on November 16, 2026.
- 02Access may include model checkpoints, source code, training and evaluation data, and monitoring systems.
- 03METR will examine what OpenAI calls its largest reinforcement-learning run to date for misaligned behavior.
- 04Transluce plans an eight-week follow-on assessment with broader model access in 2027.
- 05Assessors may publish findings, subject to OpenAI redacting confidential information.
METR's first assessment covers what OpenAI describes as its largest reinforcement-learning run to date, looking for misaligned behavior that existing tests and monitors may have missed. A second METR assessment, focused on whether OpenAI's systems catch dangerous or unintended behavior, has been preliminarily agreed, but its scope is not final.
Transluce will audit OpenAI's post-training alignment evaluations, assessing their quality and coverage, writing new evaluations to fill gaps and recommending improvements. An eight-week follow-on assessment with broader model access is planned for 2027.
Goodfire will embed interpretability researchers who use its platform to analyze model internals for training-run monitoring, debugging and alignment work. Goodfire, a public benefit corporation founded by researchers who worked at OpenAI and Google DeepMind, has raised more than $200 million, most recently at a $1.25 billion valuation.
Assessors may publish their findings, though OpenAI can redact confidential details. Apollo Research and Goodfire are paid for their work, with payment not tied to results. Transluce takes no compensation, and METR takes no payment but may accept API credits or compute.
Disputes between OpenAI and the assessors can be escalated to the Safety and Security Committee of the OpenAI Foundation's board. OpenAI said it expects this first phase to lead to longer-term embedded engagements over time.
Not every outside group is satisfied with the terms. Buck Shlegeris, CEO of Redwood Research, called the outcome "disappointing but not surprising," while Apollo Research described its agreement as "an ambitious first step towards deeper access and disclosure rights."
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.




