AIToday
Large Language ModelsAI Safety & AlignmentOpenAI BlogPublished: Nov 17, 2026, 04:00 JST

OpenAI opens models to Apollo Research, METR, Transluce, Goodfire

Send on LINE
OpenAI opens models to Apollo Research, METR, Transluce, Goodfire

3 Key Points

  1. What happened

    OpenAI signed agreements with four independent assessors — Apollo Research, METR, Transluce, and Goodfire — for a first phase of embedded assessments, giving them access to model checkpoints, source code, training and evaluation data, and monitoring systems.

  2. Why it matters

    The access lets outside experts probe alignment and safeguards over an extended period and surface failures OpenAI may not have anticipated, with public reporting meant to give people a clearer view of how risks are managed.

WHO IT HITSOpenAI's safety and policy teams, and the outside assessors themselves, gain structured access and publication rights; the findings are aimed at internal OpenAI researchers and oversight bodies, with public reports intended to inform how outsiders judge the lab's safety work.

Not sure about something? Ask the AI

Questions and answers are published on this page.

Summaries like this, every morning.

Context & Analysis

OpenAI frames the program as a step beyond its earlier third-party evaluation work. The company says the embedded assessments build on years of work with outside evaluators, including a recent pilot assessment of its internal monitoring systems with deep internal access and an investigation of the Hugging Face incident. It also ties the effort to commitments made under the White House Accord on Super Intelligence, where OpenAI agreed to independent assessments of its safety controls.

The protocols described suggest OpenAI is trying to balance openness with control. Assessors needing sensitive confidential information work on site at OpenAI offices on OpenAI devices, and are expected to follow at minimum the same security standards as OpenAI employees. Where direct access could expose its most sensitive IP, such as model weights or architecture, OpenAI says it will consider alternatives, including pairing assessors with its own researchers to run analyses. Discussions with employees are off the record, though a separate attestation pathway lets assessors request OpenAI-attested statements.

Each assessor has a distinct remit. Goodfire will embed interpretability researchers using its platform to analyze model internals, aimed primarily at internal OpenAI researchers. METR is intended to conduct two assessments, one on misaligned behavior in OpenAI's largest reinforcement learning run to date and another monitor assessment that OpenAI says is pending mutual agreement on scope. Transluce will examine the quality and coverage of post-training alignment evaluations. OpenAI says it expects this first phase to lead to longer-term, consistent embedded engagements, and that it is exploring future assessments of compliance with its safety frameworks and of its organizational culture and processes.

FAQ
What exactly do the embedded assessors get to see?
Depending on each evaluation's needs, assessors receive access to some of the same internal resources OpenAI researchers use — model checkpoints, source code, training and evaluation data, monitoring systems, technical documentation, and discussions with the responsible researchers and engineers.
What is Apollo Research assessing?
Apollo Research will assess whether models could covertly work against their developers' intentions and whether OpenAI's safeguards would detect such behavior, testing four claims in two stages, including that training did not incentivize subversion and that reasoning behind subversion would be detectable.
Can OpenAI block publication of the assessors' findings?
Assessors maintain independence in publishing their assessments, but OpenAI retains the right to redact confidential information and may request responsible disclosure delays to act on findings that could pose security or safety risks and remediate issues that remain.
Who oversees these assessments inside OpenAI?
The Safety and Security Committee (SSC), an independent oversight committee of the OpenAI Foundation's Board, will have visibility into the embedded assessment and receive unredacted reports, and unresolved disagreements are referred to the SSC.
Send on LINE

Also reported by TechCrunch AI

AI news that matters for your work, delivered every morning.

Pick your industry and the AI tools you use, and get news related to your work every day.

Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →

Ask AI

Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.

Questions and answers are published on this page.

Related Articles

Next articleUS moves to close 'remote access' chip loophole