OpenAI is bringing independent AI safety researchers in-house, folding a slice of the outside scrutiny ecosystem into the company it is meant to watch.

The timing matters. Independent safety work has expanded dramatically over the past two years — university labs, nonprofit evaluators, small research shops and grant-funded individual researchers — largely because the largest labs cannot be the sole judge of their own systems. That outside ring exists to hold the inside accountable, and its people are exactly the talent an in-house team wants to hire.

Recruiting from it cuts both ways. The lab gains researchers who have spent their careers thinking about failure modes, evaluation design and disclosure norms from the outside. The independent ecosystem loses some of the voices that made it credible, and the harder question is whether those researchers keep publishing externally once their paycheques come from the lab.

It follows a turbulent stretch for OpenAI's safety organisation: in October the company parted ways with three safety researchers over data shared with an outside evaluator, and the EU opened an investigation into an unreported AI safety incident. Against that backdrop, absorbing outside talent can read as a genuine expansion of capacity — or as a way to move scrutiny indoors where it is easier to manage.

Both readings are plausible; which one holds will depend on what the newly internal researchers are allowed to say in public.