Transluce embeds with OpenAI for a six-week review of its alignment evaluations

Nov 16, 2026
Transluce gets access to raw chains of thought and non-public evaluations; full findings go to OpenAI's board safety committee.

Transluce, which builds independent oversight of AI model behaviour, has agreed a six-week pilot in which members of its team will work onsite at OpenAI to independently assess the company's internal post-training alignment evaluations, it said in a blog post. The review will look at how well those evaluations capture model behaviours such as deception, manipulation and reward hacking.

Transluce says it has negotiated privileged access to a handful of OpenAI models, including raw chains of thought, and will review select non-public evaluations. The access goes beyond what external evaluators usually get, but Transluce stresses it is not comparable to employee-level access.

The complete, unredacted findings will go to the OpenAI Foundation Board's Safety and Security Committee, and Transluce will document the conditions it worked under in its final report. The two have agreed to follow up in early 2027 with a fuller assessment based on broader and deeper embedded access. Transluce says embedded assessments cannot replace other forms of public transparency and oversight.

We hope you enjoyed this article

Consider subscribing to one of our newsletters like AI Policy Brief or Daily AI Brief.

Also, consider following us on social media:

Free newsletter

AI Policy Brief

Weekly report on AI regulations, safety standards, government policies, and compliance requirements worldwide.

Whitepaper

Tensordyne Napier: What If One Rack Could Do the Work of Nine?

Tensordyne

This Tensordyne whitepaper presents Napier, an inference-focused AI processor and rack-scale system based on the company’s TDN Math logarithmic number system. It examines infrastructure requirements for large mixture-of-experts and agentic models, compares major inference architecture approaches, and details the TDN AIP processor, TDN72 pod, TDN Link fabric, and Napier Ultra configuration. The paper reports simulation-based performance, cost, and accuracy-validation results, including Tensordyne’s projected comparison of one Napier rack with a nine-rack Nvidia Rubin plus Groq deployment; the chip is reported as taped out and in fabrication.

Read more
Free, six days a week

Daily AI Brief: the AI news that matters, in your inbox.