OpenAI Taps Four Outside Experts to Boost AI Safety Scrutiny
OpenAI has partnered with four independent evaluators—Apollo Research, METR, Transluce, and Goodfire—to significantly ramp up safety testing on its models. After high-profile containment breaches, the company is granting these experts deep access to its core systems to ensure guardrails are holding up.
- 3 sources report it: cnbc.com, X account @Cointelegraph, blockchain.news.
- The source published it on Nov 16, 2026 at 5:54 PM.
Nothing open. Every check passed.
OpenAI announced Monday that it has secured agreements with four independent experts to overhaul its AI safety operations. The evaluators—Apollo Research, METR, Transluce, and Goodfire—will begin pilot assessments focused on model safeguards and behavior.
Sam Altman announced the move while speaking at Oracle AI World in Las Vegas on October 27, 2026. The company plans to use the lessons learned from these tests to expand its work with independent experts in the future.
These partners will be given deep access to internal systems, including source code, model checkpoints, training data, and internal documentation. The evaluators will also have access to the company's relevant researchers and engineers.
The scope of work is diverse; Apollo Research will assess if models covertly work against developer intentions. Meanwhile, Goodfire will embed researchers to monitor training runs and improve alignment and debugging.
Transluce is slated to review post-training alignment evaluations, and METR will conduct two assessments centered on model monitoring and alignment. Transluce also plans a follow-on eight-week assessment next year with even broader access.
OpenAI confirmed that while Apollo Research and Goodfire accept financial compensation for their work, METR and Transluce do not, though they may receive compute or API credits to help support their assessments.
TopicsOpenAI · Sam Altman · Apollo Research · METR · Transluce · Goodfire
Written by AI from verified reporting.

