The person who wrote OpenAI's safety rulebook just quit OpenAI

The institutions deciding when frontier AI is safe to release are losing the people who built the gate.

David Robinson resigned from OpenAI last week. The number on the wall behind him is institutional: he oversaw the safety reports for 12 frontier launches and helped draft the Preparedness Framework, the document that decides when a model is too dangerous to ship. The wire read this as another insider-quits story. The actual story is that the drafter left the gate, and the gate still has to work.

Robinson's ask is concrete: run AI labs like nuclear power plants, with layers of redundancy and slow planning, and demand something close to perfection the first time on the most capable releases. The leading alternative is already in motion. Dario Amodei has called for third-party evaluators and a slower frontier cadence. Sam Altman has publicly endorsed that plan. The blueprint is named, signed by competing CEOs, and waiting on adoption.

The contest is no longer over whether to build safety scaffolding. It is over who runs the review, what it costs, and which labs adopt it before the next launch.