Caitlin Kalinowski resigned from OpenAI on Saturday — and her exit reveals a fracture inside AI's biggest labs that nobody fully planned for.
Kalinowski led OpenAI's robotics team. Before that, she built Meta's augmented reality glasses — the Orion project. She joined OpenAI in November 2024. On March 7, 2026, she walked out. The reason: OpenAI's deal to deploy its AI models inside the Pentagon's classified networks.
"Surveillance of Americans without judicial oversight and lethal autonomy without human authorization are lines that deserved more deliberation than they got," she wrote on LinkedIn.
OpenAI clarified quickly. "No domestic surveillance and no autonomous weapons" are its stated red lines. A spokesperson confirmed her departure and said the company "will continue to engage in discussion with employees, government, civil society and communities around the world."
Kalinowski's own follow-up: "My issue is that the announcement was rushed without the guardrails defined. It's a governance concern first and foremost."
So it's not about the end goal. It's about how fast they got there.
How We Got Here
The backstory matters.
A few weeks ago, the Pentagon was in talks with Anthropic — not OpenAI — to deploy AI in classified environments. Anthropic tried to negotiate hard limits: no mass domestic surveillance, no fully autonomous lethal systems. The Pentagon said no.
Instead of accepting Anthropic's terms, the Department of Defense designated Anthropic a "supply-chain risk" — a label that effectively bars it from sensitive government contracts. Anthropic is fighting the designation in court. Meanwhile, Microsoft, Google, and Amazon have all confirmed they'll keep Claude available to non-defense customers.
OpenAI saw the opening. Within days of Anthropic's fall-out, Sam Altman announced OpenAI's own Pentagon agreement — with what the company describes as "technical safeguards" for responsible national security use. The deal covers deployment in classified environments.
The speed of that move is exactly what Kalinowski flagged.
Why This Resignation Is Different
High-profile AI ethics departures aren't new. But this one cuts differently because of who resigned and what she didn't say.
Kalinowski explicitly did not say OpenAI's deal is wrong. She said the process was wrong. She said she has "deep respect" for Sam Altman. She's not sounding an alarm about OpenAI becoming a weapons company. She's saying: for decisions this consequential, the internal deliberation should match the stakes.
That framing is harder to dismiss than a blanket "AI in the military is bad" take. It's an institutional critique.
It also signals something about OpenAI's culture under pressure. The Anthropic-Pentagon fallout created a competitive vacuum. OpenAI moved fast to fill it. Fast enough that a senior leader — someone who built hardware for Meta and cared enough to join OpenAI specifically for its mission — felt the process broke down.
The Real Split in AI
What this week exposed isn't OpenAI vs. Anthropic. It's a fundamental question about what AI labs owe their own people when taking on military contracts.
Anthropic tried to hold the line on safeguards and got blacklisted for it. OpenAI took the deal and lost a senior exec over the speed of the decision. Both labs now sit in complicated positions — Anthropic fighting a government label in court, OpenAI managing internal fallout from its Pentagon deal.
The companies trying to dodge this question — by either refusing all defense work or accepting all of it with no internal guardrails — are living on borrowed time. The US government is not going to stop wanting AI for national security. And the people building that AI are not going to stop having opinions about how it's used.
What Builders Should Watch
Three things to track from here:
1. OpenAI's internal policy process. Kalinowski's criticism was governance, not principle. If OpenAI responds by formalizing how major deals get reviewed internally, this resignation will have done real work. If they don't, the next one won't come with a respectful LinkedIn post.
2. The Anthropic court case. "Supply-chain risk" is a legally specific designation with serious consequences. If Anthropic wins, it sets a precedent: labs can negotiate safeguards without facing government retaliation. If Anthropic loses, every other lab gets the message.
3. The next deal. OpenAI's Pentagon partnership is one. Google, Meta, and others have their own government contracts and ambitions. The lines being drawn right now — between acceptable national security AI use and autonomous weapons — will shape every one of those deals.
Kalinowski said this was "about principle, not people." That distinction is going to matter a lot in the next year.
Subscribe to NeuralWire for daily AI signal — the news that actually moves the needle. →
FOLLOW NeuralWire on X for daily AI signal — what matters, why it matters, what to do about it. →