OpenAI commissioned METR to review the largest scale reinforcement learning training

Nov 17, 2026 00:26:34

OpenAI has commissioned the AI safety assessment organization METR to conduct a safety review of its largest reinforcement learning training run to date. This review is part of a single-run pilot evaluation, and METR will receive model snapshots as well as internal records from the training process.

According to The Defiant, the two parties have not yet reached an agreement on a broader scope for monitoring evaluations, and the scope of the relevant assessments is still to be determined. METR, formerly known as ARC Evals, has long provided independent pre-deployment evaluations for cutting-edge AI laboratories and previously conducted red team testing on OpenAI's internal misalignment monitoring system.

Recent Fundraising

More
Nov 13
$50MNov 10
$11MNov 10

New Tokens

More
Dec 31
Dec 31
SSaturnSTRN
Dec 31

Latest Updates on 𝕏

More
SSolanaFollowedRiftATM
Nov 14
Nov 14