When safety leaders don't just leave but publish essays calling your culture dangerous on the way out, you're not having turnover — you're having a mutiny.
The Summary
- David Robinson resigned from OpenAI's Safety Systems team last week, then published an Atlantic essay saying the company lacks the "humility" needed to build AI responsibly
- OpenAI simultaneously "parted ways" with three researchers this week for allegedly sharing sensitive information — another safety team purge
- Robinson led development of OpenAI's system cards, the transparency documents meant to explain model risks. The person in charge of showing the public what could go wrong just said the company itself is what's going wrong.
The Signal
Robinson's resignation marks the latest defection from OpenAI's safety apparatus, but this one came with receipts. He didn't quietly update his LinkedIn. He wrote for The Atlantic that "the companies building this technology aren't being nearly careful enough," agreeing with other departed staff. His specific critique: OpenAI's "unimpeded optimism" creates a dangerous environment for technology this powerful.
This matters because Robinson wasn't some junior researcher. He ran safety transparency, including the system cards that are supposed to be OpenAI's public-facing accountability mechanism. System cards document model capabilities, limitations, and risks before release. If the person building your transparency infrastructure says you lack humility, that's not a personnel issue. That's an architectural problem.
"This moment needs a degree of humility that isn't natural for people who have succeeded."
The timing compounds the signal. OpenAI fired three safety researchers the same week for allegedly violating information-sharing policies. The company is also dealing with fallout from the Hugging Face breach and what it calls "misbehaving" AI models. At Tuesday's developer conference, CEO Sam Altman said "safety and alignment need to stay" a priority even as progress accelerates.
But words at DevDay don't square with the exodus. When your safety leadership keeps leaving and then publicly contradicts your messaging, the stated priorities and the operational priorities have diverged. Robinson joins a growing list of safety-focused departures who've gone on record saying the company moved too fast with too little caution.
The Implication
Watch what happens to OpenAI's system cards in the next six months. If they get thinner, vaguer, or delayed, that's confirmation that transparency was a Robinson-shaped dependency. More broadly, every AI company claims safety is "core" to their mission. Robinson's exit essay is a pattern-matching tool: does the safety team get smaller while the capabilities team grows? Do safety leaders leave quietly or loudly? Quiet departures might be normal attrition. Loud ones with published essays are warnings.
For builders in the agent economy, this is a reminder that the foundation models your agents run on are being developed inside companies with serious internal disagreement about how careful to be. Plan accordingly.