OpenAI's head of hardware resigned Saturday over the company's deal to deploy its AI inside the Pentagon's classified cloud — and her parting words cut straight to the core tension shaping AI's military future.
Caitlin Kalinowski, who led OpenAI's robotics and hardware teams, announced her resignation on LinkedIn and X on March 7, 2026. She made clear the decision was not about personalities — "I have deep respect for Sam Altman and the team" — but about governance. "AI has an important role in national security," she wrote. "But surveillance of Americans without judicial oversight and lethal autonomy without human authorization are lines that deserved more deliberation than they got."
What Happened
OpenAI announced its Pentagon deal on February 28 — less than 24 hours after Anthropic walked away from a similar agreement because the Department of Defense refused to guarantee safeguards against mass domestic surveillance and fully autonomous weapons. The Pentagon then designated Anthropic a "supply-chain risk," a label Anthropic is fighting in court.
OpenAI stepped in quickly. The company said its agreement "creates a workable path for responsible national security uses of AI" and drew its own red lines: no domestic surveillance, no autonomous weapons. But Kalinowski's point — echoed by others watching the sequence play out — is that those guardrails were announced with the deal, not established before it. In a follow-up post on X, she wrote: "My issue is that the announcement was rushed without the guardrails defined. It's a governance concern first and foremost. These are too important for deals or announcements to be rushed."
An OpenAI spokesperson confirmed her departure to TechCrunch.
Why This Matters
Kalinowski isn't a peripheral figure. She joined OpenAI in November 2024 after leading Meta's augmented reality hardware program — she was the person behind Orion, Meta's first true AR glasses. Building physical AI infrastructure is one of OpenAI's declared strategic priorities. Losing her signals internal friction that goes beyond a single disgruntled employee.
The deeper issue is the governance gap she named. When Anthropic tried to negotiate safeguards into its Pentagon agreement, the DoD said no. When OpenAI announced its deal, the safeguards appeared in a press release, not a binding contract. Whether those commitments hold — who enforces them, what the consequences are for violations — remains undefined. That vagueness is exactly what Kalinowski called out.
This is also the second high-profile departure from OpenAI tied to its relationship with the U.S. government in under a year. The pattern is notable: the people building the physical layer of OpenAI's future — hardware, robotics, embodied AI — appear to have a higher discomfort threshold for military use than the company's leadership.
The Broader Picture
The Pentagon-vs-AI-labs standoff is becoming a defining fault line in 2026. Three dynamics are now locked in:
1. The DoD wants frontier AI, on its own terms. The Pentagon's designation of Anthropic as a supply-chain risk was a warning shot. It told every AI lab: negotiate on our terms or face procurement exclusion. OpenAI accepted. Anthropic refused. Google, Microsoft, and Amazon said they'll continue offering Claude to non-defense customers, but the message landed.
2. Labs are splitting internally on where the line is. Safety commitments that look clear from the outside are fuzzy from the inside. Kalinowski's concern wasn't that OpenAI signed with the Pentagon — it was that the governance framework wasn't ready when the announcement dropped. That's a process problem with large consequences.
3. The Pro-Human Declaration is gaining traction. A cross-industry document finalized before the Pentagon-Anthropic standoff — calling for binding limits on AI in surveillance and weapons — now has renewed relevance. The declaration has no enforcement mechanism, but it's becoming a reference point for internal dissent.
What Builders Should Watch
If you're building on OpenAI's API or planning to integrate its models into products, none of this changes your immediate roadmap. OpenAI's consumer and enterprise services are unaffected.
But the larger signal matters: the same pressure the DoD applied to Anthropic can be applied to any lab that holds a significant government relationship. If you're building for regulated industries — healthcare, finance, defense contractors — the question of whether your AI vendor has binding, auditable usage constraints is moving from "nice to have" to "due diligence requirement."
Kalinowski's resignation won't stop OpenAI's Pentagon work. What it does is put a name and a face to the cost of moving fast on governance questions that don't have easy answers. In a sector where the default move is "ship first, figure it out later," that's worth paying attention to.
FOLLOW NeuralWire on X for daily AI signal — what matters, why it matters, what to do about it. →