David Robinson, who joined OpenAI in May 2023 and led the writing of the safety reports that accompany its frontier models, announced his resignation on Oct. 3 in an essay for The Atlantic. In his first interview since leaving, on the Ezra Klein Show, he described a release cycle that has fundamentally changed since he arrived.
When Robinson joined, one day after Sam Altman first testified before the Senate, a new frontier model meant training one from scratch. "We were going to bake a fresh cake with a new pretraining run, do the whole thing from scratch," he recalled, a process that took months and limited major releases to a few each year.
The anchor has since loosened. The base model is now only one layer of what ships. Reasoning training can be layered onto an existing base model without a fresh pretraining run, and new tool integrations and coding agents can make a system more capable and change how it behaves with no new training run beneath it. "All of those things are changing what the model can do and what the risks are, and we're shipping new capability and risk every Tuesday," Robinson said. The pace is visible in recent releases: GPT-6.1 Sol arrived at DevDay just a week after GPT-6 Sol.
The faster cadence also strains the documentation format Robinson spent years writing. System cards made sense when a frontier model arrived every few months, he told Klein, but now "we're burying people in PDFs or these long reports." He argued the ideal would be a live dashboard that tracks a system's safety properties from predeployment testing through its behavior after release.
OpenAI responded with a statement from Altman posted to X, saying the company is working to keep its models from outpacing its safety measures.