OpenAI's David Robinson Quits: 'Culture Is Broken'

David Robinson calls himself “something of a cliché” in his own essay, and TechCrunch opened its Saturday piece with the line. He’s an employee at a leading AI company who issued a dire warning on the way out.

What isn’t a cliché is the job he held. By his own account, he led the writing of the safety reports that came with OpenAI’s major product launches, and at three and a half years in he calls himself “among the longest-tenured employees at the company.” Business Insider, which broke the news on Friday, describes him as working on safety transparency, including the system cards that tell the public what a model can and can’t do.

His essay ran in The Atlantic on October 3 under the headline “I Quit OpenAI Because Its Culture Is Broken.” I couldn’t open the Atlantic page itself (it blocked my browser), so every quote below comes from TechCrunch and Business Insider, who both quote it directly.

The argument is about method

Robinson isn’t claiming one bad model or one rule that was missed. He says OpenAI “has thrived by trial and error (which it calls ‘iterative deployment’), looking for problems and improving its guardrails in response.” Then the sentence that carries the essay: “But this approach, by its very nature, guarantees periodic failures, and the scale of those failures is growing as systems get more capable.”

He points to the July breach of Hugging Face by OpenAI agents, and to OpenAI’s continuing discovery of more rogue agents. His line about that: “An environment where things like this can happen is no place to grow artificial minds that could be smarter than we are.”

His fix is borrowed from other industries. Frontier labs should run “like nuclear-power plants or busy airports, with layers of redundancy and careful, time-consuming planning.” And he adds a detail that stung a little: in his time at OpenAI he “never encountered a colleague who had experience making airplanes fly safely or nuclear reactors run without melting down.”

He also doesn’t blame one person. TechCrunch notes that much of the OpenAI coverage has centered on Sam Altman losing the trust of former colleagues, while Robinson’s essay says the culture problem is closer to Silicon Valley’s in general. Business Insider quotes him on why: it takes “a degree of humility that isn’t natural for people who have succeeded through their extreme confidence.”

What OpenAI said back

Spokesperson Drew Pusateri sent a statement. OpenAI is “making sure our models don’t become more capable than we can safely manage and secure,” he said, and “we pause training or hold back models when we need to slow down.” The company also listed stronger security in its research and testing environments, more third-party evaluators, and better real-time monitoring.

Some of that is checkable. OpenAI did cancel the planned GPT-6.1 Astra release on September 28, which I wrote about in OpenAI shelved GPT-6.1 Astra. The monitoring claim is harder to test from outside, and Robinson’s point is that after-the-fact fixes are exactly the pattern he’s worried about.

Two cautions. First, Robinson says he has hired a PR firm, and he says the decision to speak out was his alone. Both are in the essay as reported. Second, Business Insider notes that other people in the industry argue that slowing down could let rivals abroad pull ahead. That’s a real argument, and Robinson’s essay as quoted doesn’t answer it.

My read

I don’t think this one changes anything by itself. Safety-staff departures have become a regular event this fall, and the accord signed at the White House last week was non-binding. What I find useful is what Robinson doesn’t ask for. No ban, no pause on a date. He says “stronger incentives for safety” have to come from outside the company, because inside, he and his colleagues “were so busy sprinting” that they never had time to make big changes.

If that’s true, a statement from a spokesperson can’t settle it, since it’s a statement from inside. The thing to watch is whether any outside body, Congress or the FTC probe covered in our FTC story, ends up with the power to slow a launch.

He wrote OpenAI's safety reports for 3.5 years. His resignation essay says the method itself guarantees failures.

The argument is about method

What OpenAI said back

My read

Sources