Caitlin Kalinowski, who led OpenAI’s hardware and robotics team, resigned on Saturday over the company’s deal to deploy its AI inside the Pentagon’s classified cloud networks. Her departure is the most public internal break yet over OpenAI’s pivot toward military applications.
What happened
Kalinowski announced her resignation on LinkedIn, writing: “AI has an important role in national security. But surveillance of Americans without judicial oversight and lethal autonomy without human authorization are lines that deserved more deliberation than they got.”
In a follow-up post on X, she sharpened her critique: “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.”
OpenAI confirmed her departure. The company said: “We believe our agreement with the Pentagon creates a workable path for responsible national security uses of AI while making clear our red lines: no domestic surveillance and no autonomous weapons.”
Kalinowski, who previously led the hardware team for Meta’s Orion AR glasses, joined OpenAI in November 2024.
The backdrop: Anthropic said no. OpenAI said yes.
This deal stems from a breakdown between the Pentagon and Anthropic. Anthropic walked away from negotiations when the DoD refused to define specific safeguards against mass domestic surveillance and fully autonomous weapons systems. The Pentagon responded by formally designating Anthropic a supply-chain risk — an extraordinary move that effectively bars government reliance on the company.
Anthropic says it will fight the designation in court. Microsoft, Google, and Amazon confirmed they will keep Claude available to non-defense customers in the meantime.
OpenAI moved quickly. Sam Altman announced a deal with the Pentagon on February 28th, allowing its models to operate in classified environments. The company says it has defined “red lines” — but those limits were not publicly outlined before the announcement. That is exactly what Kalinowski objects to: governance after the deal is weaker than governance before it.
Why this matters
The AI industry is splitting. One side: companies willing to work with defense, betting that structured military partnerships beat a world where AI gets deployed with no commercial input at all. OpenAI is making that bet. The other side: companies — and now individual executives at the highest level — who believe crossing certain lines without pre-defined guardrails creates more risk than it prevents.
Kalinowski is not anti-military AI. She is anti-rushing governance to close a deal. When Anthropic asked for specific protections, the Pentagon refused. OpenAI accepted without those protections locked in. The gap between “we have defined red lines” (OpenAI’s statement) and “guardrails were not defined before the announcement” (Kalinowski’s complaint) suggests the internal clarity may not match the public positioning.
This is also the second high-profile hardware exit for Kalinowski. She left Meta after leading Orion — widely considered the most technically advanced AR glasses hardware of 2024. Her departure from OpenAI after 16 months signals that the company’s direction has shifted in ways she could not stay for.
What’s next
The core question: will the guardrails Kalinowski asked for — specific limits on surveillance and autonomous lethal systems — get codified into the Pentagon agreement, or stay aspirational? OpenAI’s statement that it will “continue to engage in discussion” is deliberately vague.
For builders and researchers watching from outside: the AI safety debate has moved from abstract to concrete. Not “should AI be aligned” but “should your specific model run on classified military networks without pre-defined limits.” The pressure from this resignation may force OpenAI to publish the specifics of its red lines before they get tested. If those lines hold, this is a meaningful precedent. If they bend under operational pressure, we will know the governance was theater.
Subscribe to NeuralWire for daily AI signal → neuralwire.io
FOLLOW NeuralWire on X for daily AI signal — what matters, why it matters, what to do about it. →