OpenAI Safety Lead Quits, Blasts AI Industry Culture

David Robinson spent years writing the safety reports attached to every major OpenAI model release, the documents meant to tell the public what a new system could and couldn't do before it shipped. This week he quit, and instead of a quiet exit, he went to The Atlantic to explain why.

The timing matters. OpenAI, like most of its rivals, has spent the past few years moving fast, shipping bigger models on tighter schedules while regulators and outside researchers scramble to keep pace. Robinson was inside that machine, producing the paperwork that's supposed to reassure everyone it's under control.

Now he's outside it, and he's not describing a few policy gaps that better rules could patch. He's describing something closer to a cultural problem, baked into how Silicon Valley builds things in general and how frontier AI labs build things in particular.

It's fair to be skeptical when the people who built the thing start warning about it after they've already cashed out or moved on. But Robinson's specific position, writing the exact documents meant to catch these problems, gives his account a different kind of weight. Here's what he told The Atlantic.