So here's a thing that happened over the weekend: Sam Altman, the CEO of OpenAI (MSFT), announced that his company is now willing to work on classified projects for the Department of War. This is a pretty big deal because, until now, OpenAI had been sticking to unclassified stuff. Altman described the shift as urgent and said it's "far more complicated" than earlier efforts.
Think of it like this: OpenAI had been the kid at the party who only wanted to play board games in the living room. Now they're saying, "Actually, I'll go into the basement where the adults are talking about serious stuff." The basement, in this metaphor, is classified military work.
The change comes after OpenAI reached an arrangement with the Pentagon that includes two specific guardrails: no domestic mass surveillance and human control over any use of force. Altman said the company had been planning to stick to non-classified work for a long time and had even turned down classified opportunities that its rival, Anthropic, accepted.
Why the Rush?
Talks with the Department of War on non-classified work had been going on for months, Altman said, but the classified track "accelerated sharply" during the week. He framed the decision as support for a mission he called critical, while arguing that the government "should not be outmuscled by private executives."
The Pentagon deal isn't just policy language; it includes practical steps like placing OpenAI engineers on-site to monitor model behavior and safety. Altman also said OpenAI will build technical controls to keep systems operating within expected bounds, and that the Department of War wanted those protections too.
Here's where the timing gets interesting. This announcement landed within hours of a major break between Washington and Anthropic. The Trump administration blacklisted Anthropic after a dispute tied to the same two restrictions that OpenAI says the Pentagon accepted in its own deal.
The Anthropic Parallel
Anthropic's Claude AI had already reached classified military networks under a contract that could run up to $200 million. But the relationship soured when the Pentagon pushed to delete contractual limits tied to surveillance of Americans and autonomous weapons use. The department said it needed freedom to deploy the system for all lawful uses, even while stating it had not sought those contested applications.
When Anthropic wouldn't budge, Defense Secretary Pete Hegseth tagged the company as a supply chain risk, and President Donald Trump directed federal agencies and military contractors to sever ties with it. Anthropic responded Friday that it was "deeply saddened" and said it would fight the designation in court, calling it "legally unsound" and warning it would "set a dangerous precedent for any American company that negotiates with the government."
Altman said the rush on OpenAI's side was meant to cool down what he viewed as a dangerous trajectory for Anthropic, for competition among AI labs, and for the U.S. as a whole. He also said OpenAI negotiated so that comparable terms would be available to other AI developers, not just his company.
The Two Red Lines
The two conditions at the center of OpenAI's Pentagon work mirror the lines Anthropic says it drew: a ban on domestic mass surveillance and a requirement that humans retain control over decisions involving force, including autonomous weapons systems. Altman also said the Department of War viewed those principles as consistent with existing U.S. law and policy.
So here's the puzzle: both companies publicly described nearly identical constraints. OpenAI says it secured acceptance of the guardrails, while Anthropic ended up blacklisted. The unresolved question is what OpenAI agreed to that Anthropic didn't. Was it the on-site engineers? The technical controls? Something in the fine print? We don't know yet, but it's the kind of detail that makes all the difference in government contracting.
It's a classic case of two companies walking up to the same line in the sand, but only one of them figuring out how to cross it without getting their feet wet. Or, in this case, without getting blacklisted.
