David Robinson, who says he led the writing of safety reports accompanying OpenAI’s major launches, has resigned after three and a half years at the company. He objects to a working culture built around successive sprints that, in his judgment, leave too little room to build the safeguards increasingly capable AI systems require.
In an essay reported by TechCrunch on October 3, Robinson described OpenAI’s culture as “broken.” That is a former employee’s assessment, not an independently established finding. His account raises a specific operational question: can a company organized around rapid iteration reliably contain systems whose failures may become more consequential as their capabilities grow?
OpenAI says it is strengthening its security and safety practices, including holding back models and pausing work when necessary. Its disclosures document both control failures and corrective action, giving the disagreement a concrete basis beyond differing views on how quickly AI development should move.
Robinson’s Criticism Targets the Launch Cycle
Safety reports are where a company explains a model’s risks, testing and safeguards to the public. According to Robinson’s account, he led that writing for OpenAI’s major product launches. He was involved in communicating the company’s safety work, giving his criticism a perspective unavailable to an outside observer relying solely on product behavior.
There are limits to what that role establishes. Leading safety-report writing does not, by itself, show what authority Robinson held over release decisions or make his interpretation of the entire organization definitive.
His central argument is that OpenAI’s approach to learning through deployment becomes less acceptable as systems become more capable. He wrote:
“OpenAI has thrived by trial and error (which it calls ‘iterative deployment’), looking for problems and improving its guardrails in response.”
Robinson argues that this approach “guarantees periodic failures” and that the scale of those failures is growing. This is his contention about the operating model’s consequences, not a measured finding that every release creates unacceptable risk.
The Guardian’s account of his essay captures his cultural criticism directly: “As the company sprints from one launch to the next, it is failing to achieve the level of care that I believe is needed.”
