OpenAI wants a team assigned to argue its next model is dangerous.
OpenAI published a framework this week for proving a frontier training run is safe before it continues. Other teams draft dissent scenarios, executives hold a veto, and training pauses if the argument fails. The company calls this an aspiration rather than current practice, and limits it to reinforcement learning.
Why it mattersLabs usually check a model before release, not while it is still learning. Moving the decision earlier means someone can stop a model before anyone has shipped it.
Read in the app — free, in 9 languages
Related stories
A free Chinese model wrote exploits for holes nobody had reported.2026-09-30More than a fifth of reviewers used AI after being told not to.
2026-09-29Microsoft ran 1,024 agents on one job with nobody in charge.
2026-09-29An AI reran 4,452 economics papers and flagged most of them.
2026-09-29A wrong AI summary cut what people remembered almost in half.
2026-09-29