Summary: Caitlin Kalinowski, OpenAI’s head of robotics and consumer hardware, resigned on March 7, 2026, citing the Pentagon agreement. She said AI has a national-security role but objected that the announcement came before surveillance and lethal-autonomy guardrails were defined and verifiable. Consumer backlash included a 295% surge in ChatGPT uninstalls, turning the governance dispute into a product-trust risk.
On Saturday, March 7, 2026, Caitlin Kalinowski — OpenAI’s head of robotics and consumer hardware — took to LinkedIn to announce she was stepping down, and she pointed straight at the company’s agreement with the Pentagon as the reason. “This wasn’t an easy call,” she wrote, describing the move as “about principle, not people” while offering “deep respect” to Sam Altman and her colleagues.
Few departures from OpenAI in recent years carry this much weight, and few land at a more loaded moment. NPR’s reporting places the resignation squarely inside a race among leading AI developers to become the U.S. military’s technology supplier. Kalinowski arrived at OpenAI in November 2024 from Meta, where she ran hardware for the Orion AR glasses, and she anchored the company’s consumer hardware and robotics ambitions.
The Substance of Her Statement
Kalinowski named two red lines with precision: “surveillance of Americans without judicial oversight” and “lethal autonomy without human authorization.” Nothing in her post rules out AI’s role in national security — she states outright that “AI has an important role in national security.” What she rejected was the order of events: a deal put in front of the public before its guardrails were ever spelled out.
Her follow-up on X put it even more bluntly: “my issue is that the announcement was rushed without the guardrails defined,” a matter she called “a governance concern first and foremost.” So the fight is not over whether OpenAI belongs in military work; it is over whether the governance machinery existed before the commercial commitment did.
The Deal, the Timeline, and the Missing Guardrails
The sequence of events explains why this blew up. Per TechCrunch, Anthropic’s talks with the Pentagon fell apart over safeguards covering mass domestic surveillance and fully autonomous weapons; the Pentagon then labeled Anthropic a supply-chain risk, a designation Anthropic is now fighting in court. OpenAI moved into the opening, unveiling its own agreement around February 28 that opens classified environments to its technology and describes “a more expansive, multi-layered approach” spanning contractual and technical safeguards. Its red lines mirror Anthropic’s: no domestic surveillance, no autonomous weapons.
An OpenAI spokesperson confirmed her departure and stood behind the deal, calling it “a workable path for responsible national security uses of AI” and committing to keep talking with employees, government, and civil society.
Kalinowski’s critique goes straight at that seam: on what the red lines actually say, the two sides hardly differ. What separates them is that, on announcement day, none of it had hardened into mechanisms anyone can verify.
The Market Answers with Uninstalls
Public reaction outpaced anything happening inside the company. Per TechCrunch, the deal set off a consumer revolt: ChatGPT uninstalls surged 295%, and by Saturday Claude had claimed the top spot on the U.S. App Store free charts, with ChatGPT sliding to No. 2.
A product that grows on consumer trust cannot wave this off. Enterprise buyers watch the same numbers — the moment a model vendor’s military posture starts shifting App Store rankings and uninstall rates, it stops being a communications problem and becomes product risk.
A Question of Sequence, Not Personnel
Reading this as “one executive quit over the military” misses what she actually argued. Her framing is procedural, and it is about order: guardrails come first — who may authorize lethal action, how surveillance data is constrained, how compliance is audited — then the signing, then the announcement. Reality ran that list backward, so even where the slogans line up, neither employees nor users can pin down when those red lines turn into engineering and contractual constraints.
The bind is industry-wide in 2026: Anthropic holds the safeguards and gets tagged a supply-chain risk, while OpenAI wins the seat with softer commitments. At least in the short run, the market is effectively taxing governance-first strategy — and her resignation hauled that contradiction out of the boardroom and into public view.
Takeaways for AI Teams
Three things worth acting on. One: government contracts are now baseline requirements for frontier labs, so any product leaning on a single model vendor should write “vendor military and governance posture” into its risk register and keep a routing fallback ready. Two: “guardrails before announcements” is hardening into an explicit expectation on both the user and employee sides — transparency artifacts and red lines have to be verifiable, not just declared. Three: track the talent signals — losing the executive who owned the hardware and robotics roadmap is a genuine variable in OpenAI’s physical-product plans, and it hands competitors a recruiting opening.