September 23, 2026
OpenAI models to be tested earlier: auditors will join before release
On September 22, OpenAI said it would allow independent reviewers into model training, evaluation, and deployment stages. Reviews may last from weeks to several months and will not be tied to the date of a specific release.
With GPT-4, OpenAI had already given external organizations access to early model versions before deployment. The company now wants to involve auditors throughout the development cycle: to assess safeguards, model capabilities, and cases of dangerous behavior.
OpenAI is already in talks with several organizations about the first such reviews.
Source
