In 2023, OpenAI published a usage policy. It was clear. Military applications, weapons, warfare. All banned.
On February 28, 2026, they signed a deal with the US Department of Defense to put their AI inside classified military networks.
Three years. That’s how long the promise lasted.
How It Happened
The Trump administration banned Anthropic from government contracts. The reason: Anthropic refused to remove conditions that prohibited its AI from being used for mass domestic surveillance and autonomous weapons. They held the line. The Pentagon called them a “supply chain risk” for doing so.
Hours after that ban, OpenAI announced its own Pentagon deal. The timing was not subtle. Anthropic said no to $200 million and got punished. OpenAI said yes and got rewarded.
That’s the story. Everything else is details.
What OpenAI Says vs. What OpenAI Did
OpenAI announced three red lines: no mass domestic surveillance, no autonomous weapons, no use against civilians. They said their “safety stack” makes violations impossible.
But here’s the thing about red lines drawn by the company that benefits from crossing them. They’re not red lines. They’re starting positions in a negotiation.
MIT Technology Review called it “what Anthropic feared.” The Electronic Frontier Foundation called the safety language “weasel words.” A senior member of OpenAI’s own robotics team quit, saying the deal was rushed “without the guardrails defined.”
When your own people leave over a safety announcement, the announcement is not about safety.
The Users Noticed
ChatGPT uninstalls jumped 200% the day the deal was announced. People who had paid for subscriptions, who had integrated ChatGPT into their workflows, who had defended the company in arguments with skeptical friends, deleted the app.
Anthropic’s Claude hit number one in the App Store the same week.
This matters. Not because app store rankings determine the future of AI, but because it shows that a significant number of people drew their own line. They decided there are things they won’t support with their money and their data. And a military deal without clear limits was one of them.
The Honest Question
OpenAI was founded because its founders believed artificial general intelligence was coming and that it needed to be developed safely, for the benefit of all humanity. That was the founding story. That was the justification for nonprofit status, for the unusual governance structure, for every press release about responsible AI development.
A classified military contract. With the Pentagon. After explicitly banning it.
At what point does the founding story become a marketing strategy? At what point do we stop calling it a safety-focused company and start calling it what it is? A company that sells AI, to whoever will pay the most, with whatever conditions the buyer will accept.
The employees who quit already answered that question for themselves. The users who deleted the app did too.
Everyone else is still deciding.
[…] also partnered with the US military and Pentagon in ways that caused significant internal dissent. Employees who built the company have walked out over the Pentagon deal, and while this has not slowed fundraising momentum, it raises real questions about whether the […]