OpenAI's head of hardware, Caitlin Kalinowski, resigned on Saturday — citing the company's rushed Pentagon deal as a line she couldn't cross. She's not the first AI executive to walk over a military contract. She probably won't be the last.
What Happened
On March 7, Kalinowski announced her departure from OpenAI via LinkedIn, framing it explicitly as a governance protest. "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."
Kalinowski had been leading OpenAI's hardware team since November 2024, having come from Meta where she built the Orion AR glasses. By all accounts, she was a serious hire — not someone who quits lightly.
Her core objection wasn't the Pentagon partnership itself. It was the process. "My issue is that the announcement was rushed without the guardrails defined," she clarified in a follow-up post on X. "It's a governance concern first and foremost. These are too important for deals or announcements to be rushed."
The Deal That Triggered It
To understand the resignation, you need to understand the month of chaos that led to it.
In late February, talks between the Pentagon and Anthropic collapsed. The DoD wanted "any lawful" use rights for Anthropic's Claude models. Anthropic refused, specifically trying to negotiate protections against mass domestic surveillance and fully autonomous weapons. The Pentagon responded by designating Anthropic a "supply-chain risk" — a procurement blacklist that blocks federal contracts. Anthropic announced it would fight the designation in court.
OpenAI stepped into the void. On February 28, Sam Altman announced OpenAI's own Pentagon agreement, covering access to its models on classified cloud networks. The announcement came fast — and according to Kalinowski, without the ethical guardrails defined upfront.
OpenAI's official response to her resignation: "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."
Altman's red lines are the same protections Anthropic fought for — and failed to secure. The difference is Anthropic walked. OpenAI signed, then defined the limits after the fact.
Why This Is Bigger Than One Resignation
Kalinowski's departure is a signal, not just a story.
Three things are happening simultaneously in the AI-military complex right now:
1. The government is forcing the issue. The Trump administration has drafted strict new AI procurement guidelines requiring companies to grant "any lawful" use of their models for civilian government contracts. This isn't a request — it's the price of doing business with the federal government. Anthropic's supply-chain risk designation is the enforcement mechanism. It tells every other AI company: comply or be locked out.
2. AI labs are splitting into two camps. Anthropic chose to fight in court rather than sign on. OpenAI chose to sign and manage the guardrails internally. Microsoft, Google, and Amazon said they'll continue offering Claude to non-defense customers — a middle path. The industry doesn't have a unified position, and that vacuum is exactly what the Pentagon is exploiting.
3. Internal dissent is real. Kalinowski's resignation follows a pattern. In 2023, researchers left OpenAI over safety concerns. In 2024, the same thing happened at Google DeepMind after Project Nimbus. Now OpenAI's hardware lead is walking over a military contract. The people who build these systems have opinions about how they're used — and some of them are willing to leave over it.
What Builders Should Watch
If you're building on OpenAI's API, none of this changes your access tomorrow. But the trajectory matters:
OpenAI is now a defense contractor. That's not hyperbole — it's the category the company entered when it signed the Pentagon deal. Defense contractors operate under different pressures than consumer tech companies: classification requirements, export controls, government audit rights, and political exposure every time a contract gets scrutinized by Congress.
Anthropic's bet is different. By fighting the supply-chain designation in court, they're positioning themselves as the safety-first alternative for enterprises and governments that care about AI governance. That's a real market — the EU, financial institutions, healthcare systems, and governments outside the US have different risk tolerances.
The split between "AI that does what governments want" and "AI with built-in limits" is going to define the next phase of the industry. Kalinowski leaving OpenAI is one data point. The Anthropic-Pentagon standoff is another. Watch where the talent goes next.
FOLLOW NeuralWire on X for daily AI signal — what matters, why it matters, what to do about it. →