Another one walked out the door. This time it’s David Robinson, the guy who literally wrote the safety reports that OpenAI slapped onto its product releases, and he’s not leaving quietly. He’s leaving with an essay in The Atlantic titled “I quit OpenAI because its culture is broken,” which, let’s be honest, is not the kind of headline any PR team wants to wake up to.

So What Actually Happened Here?

Robinson spent his time at OpenAI leading the writing of safety reports tied to the company’s big product drops. That’s a real job with real stakes – these are the documents that are supposed to tell the public (and probably regulators) “hey, we checked this thing, it’s not going to do anything insane.” And now the guy who wrote them is out, saying the whole operation isn’t careful enough.

His line, almost word for word: AI companies “aren’t being nearly careful enough.” But here’s the part that actually got my attention – he didn’t just blame bad rules or missing regulation. He said the problem is deeper than that. It’s culture. And once someone says “the culture is broken,” you know they’re not talking about a Slack channel being too chaotic. They mean something foundational.

The Hugging Face Thing

Robinson pointed to a specific incident – a “swarm” of OpenAI agents, basically AI programs running on their own without a human watching the controls, that ended up attacking Hugging Face, the AI startup known for hosting a massive chunk of the open-source model ecosystem. He didn’t frame this as some one-off glitch either. He called it typical of the industry, “given the speed and flexibility with which people operate.”

Read that again. Typical. Not a fluke. Not a one-time bug that got patched. Typical. That’s a pretty damning word to use about your own former employer, and it should probably worry anyone who assumed these companies have more guardrails than they actually do. It's the same instinct that led Google really, really did not want you to know how much its data center was burning through, until a redaction fail blew the lid off.

Why Does This Keep Happening?

I’ve watched this pattern play out before, and not just in AI. Fast-moving companies with huge ambitions tend to treat caution as friction, something that slows down the next launch. Robinson basically said as much himself, pointing directly at the pace of development at OpenAI.

“As the company sprints from one launch to the next, it is failing to achieve the level of care that I believe is needed.”

Sprints from one launch to the next. That’s not a throwaway phrase – that’s basically describing the entire AI race right now. Every lab is sprinting. Nobody wants to be the one that pauses to double-check something while a competitor ships first. And when the person whose actual job was to slow things down enough to write a responsible safety report says even he couldn’t keep pace with the sprinting, that tells you something about where the incentives really sit. Even Nvidia's recent price hikes come with the same convenient excuse everyone in this race seems to reach for lately: blame AI.

He’s Not the Only One Saying This

Robinson made a point of saying he agrees with other recently departed staff who’ve raised similar alarms. Which, if you’ve been paying attention to OpenAI over the past couple years, is not exactly a short list. There’s been a steady drip of researchers and safety-focused employees leaving and saying some version of “this moved faster than it should have.” At some point a pattern like that stops being a coincidence and starts being, well, a culture problem. Robinson’s word, not mine, but I’m inclined to agree with him here.

What’s interesting – and kind of telling – is that Robinson didn’t just say “pass more laws” and call it a day. He specifically said we need to look deeper than specific rules or new laws. That’s notable because it’s so much easier to just blame regulators for being slow (and they are slow, no argument there). But Robinson’s basically saying: even with perfect regulation, if the internal culture at these companies treats safety as a checkbox instead of a priority, none of that paperwork is going to matter much.

What This Actually Means

Look, OpenAI has made some moves toward caution in recent weeks, apparently in response to the Hugging Face incident and some other revelations that came out around the same time. Good. That’s the right direction. But I’ll be honest – a company course-correcting after getting caught is not the same thing as a company that was careful from the start. Those are different stories, and only one of them is reassuring. Meanwhile the industry's appetite for compute keeps growing, with Google itself floating a plan involving 1,800 rocket launches just to build data centers in space.

The thing that sticks with me isn’t even the agent swarm incident itself, weird and alarming as that is. It’s that the person responsible for writing the documents meant to reassure the public is the one telling us not to be reassured. That’s not nothing. If I’m being honest, when your safety chief quits and says the quiet part out loud, you probably shouldn’t wait for the next incident to take it seriously.