Let me give you the TL;DR; of the latest news in AI.
It’s Tuesday, February 25th, 2026. Dario Amodei, co-founder and CEO of Anthropic, one of the most advanced AI companies on the planet, walks into the Pentagon for a meeting with Defense Secretary Pete Hegseth (I know, yeah, that guy).
The room is packed: Deputy Secretary, Under Secretary for R&D, Under Secretary for Acquisition, chief spokesperson, general counsel. The full varsity squad. They brought everyone.
Hegseth, to his credit, opens by praising Claude. Says it’s great. Loves the product. Really, really good AI. The best AI, some people are saying.
Then he gives Amodei until 5:01 PM Friday to hand over the keys, unrestricted access, no guardrails, no ethical limits, or face the consequences.
Amodei listens. Thanks him for his service. Reiterates Anthropic’s red lines.
And then, essentially, says no.
Before we get to the beautiful chaos that followed, let’s be clear about what Anthropic refused to do. Because “woke AI”, Hegseth’s charming description of the problem, doesn’t quite capture it.
Anthropic had two hard limits:
No AI-controlled autonomous weapons. As in: no letting Claude decide to pull the trigger with no human in the loop.
No mass domestic surveillance of American citizens.
That’s it. Those were the lines. Not “we won’t help the military.” Not “we hate America.” Just: we won’t let an AI autonomously kill people, and we won’t let you spy on your own population at machine scale.
Reasonable, right? Apparently not. Apparently this is ideological tuning. Apparently this is incompatible with American principles.
The Pentagon’s counterargument, delivered with a straight face, was essentially: legality is our responsibility, not yours. Just give us the tool and trust us.
Right. Because that’s gone so well historically.
How did we end up in a reality where a defense secretary is threatening to invoke the Defense Production Act, a law designed for wartime industrial emergencies, against a software company because they won’t remove safety features from a chatbot?
IMHO, the answer is: the AI industry built this trap itself.
For years, companies like Anthropic, OpenAI, and the rest of the zoo have been in a race to out-hype each other. Transformative. Revolutionary. Near-AGI. The most capable model ever built. More powerful than anything that came before. Buy now, futures on sale.
That narrative was commercially necessary. It attracted investment, talent, customers, and press. It also landed, word for word, in the ears of people who now genuinely believe these models can reliably operate weapons systems, run surveillance infrastructure, and make battlefield decisions in real time.
You spent years telling the world your AI is basically magic. You cannot be shocked when someone in power says: great, hand over the magic wand, and take off the safety on it.
Anthropic, to their credit, has actually been more honest than most about what their models can’t do. Dario’s core argument, that AI isn’t reliable enough yet for autonomous weapons,is the opposite of hype. It’s the one company that’s been relatively straight with the world about limitations now being punished for it.
The irony is almost too much to handle, lol.
Friday came. The deadline passed.
Trump posted on Truth Social ordering every U.S. government agency to “immediately cease” all use of Anthropic’s technology. Hegseth designated Anthropic a “Supply-Chain Risk to National Security”, a label typically reserved for Chinese companies and foreign adversaries.
Emil Michael, the undersecretary of defense for research and engineering and former Uber executive (make of that what you will), called Amodei “a liar with a God-complex” who wants “nothing more than to try to personally control the U.S. Military.”
The man who built safety guardrails against autonomous killing machines. A God-complex. Sure.
Meanwhile, and this is the part you should not miss, Elon Musk’s xAI was cleared for use in classified military systems right in the middle of all this. Convenient timing. Very organic. Nothing to see here :)
Senator Mark Warner said the quiet part loud: this might be the pretext to steer contracts to a preferred vendor whose model “a number of federal agencies have already identified as a reliability, safety, and security threat.”
That preferred vendor being, of course, the one owned by the guy who has a desk in the White House.
Here’s what I believe. Losing a $200 million Pentagon contract sounds catastrophic. It isn’t. That contract is less than 2% of Anthropic’s annual revenue. The reputational math here is completely different.
European and enterprise customers, you know the ones in healthcare, finance, legal, critical infrastructure, have been quietly terrified of American AI companies for years. Post-Schrems, post-CLOUD Act, they’ve watched the U.S. government assert ever-expanding rights over data and technology. An AI vendor that just demonstrated it will hold its ethical line even against a direct order from the U.S. government? That’s the vendor you want handling sensitive data. That’s a compliance officer’s dream. I am not saying that all the other issues of data migration and ownership are gone though (pay attention to that!)
The talent market is the other piece. The best AI researchers, the people who actually build these systems,are disproportionately the kind of people who care about this stuff. Deeply. Capitulating would have triggered a talent exodus. Holding the line is a recruiting poster. Expect Anthropic’s next hiring cycle to be exceptional.
And the narrative is just clean. “We refused to let AI kill people autonomously and spy on citizens. They blacklisted us.” That story fits on a t-shirt. It will be in the first paragraph of Anthropic’s Wikipedia page for the next twenty years.
History has a soft spot for the companies that took the hit and were right.
Friday afternoon, after Anthropic had already taken the full blast, Sam Altman published an internal memo stating that OpenAI has the same red lines. No autonomous lethal weapons. No mass surveillance. Humans in the loop.
The solidarity is appreciated. It would have been more appreciated forty-eight hours earlier.
But it matters, because it reframes the whole story. This isn’t a fringe position. This isn’t one ideologically captured company being precious about its product. The two most powerful AI labs in the world, bitter competitors in every other context, agree that some lines shouldn’t be crossed. The administration picked a fight over ethics and accidentally created a unified front.
So what do we actually learn here?
The hype cycle has consequences beyond stock prices and misallocated venture capital. When you oversell what AI can do, you invite exactly this kind of pressure from people who believed you. The industry built its own cage.
The “woke AI” framing is not confusion, it’s a weapon. Don’t mistake deliberate political theater for genuine technical misunderstanding. These people know what they’re asking for.
And when an administration that fast-tracks a politically connected competitor’s classified clearance while blacklisting the one company that said no, that’s not a technology policy. That’s clearly a shakedown. Anthropic said no. That matters. Not because they’re perfect, not because this ends cleanly, but because someone had to. And the first company to say no to this particular administration, on these particular issues, gets to own that story forever. Long story short, in five years, nobody will remember the $200M contract. Everyone will remember who held the line.
Defrag Zone is a no-hype technology newsletter for people who think carefully about where this is all going. If you’re not subscribed yet, you’re reading someone else’s copy, fix that here.