The label held. A domestic American AI company just lost a court fight over a designation built for foreign adversaries — and now it's case law. If you're just joining, here's the fight: Anthropic and the Pentagon have been arguing over whether Claude can run on classified military systems while Anthropic keeps its red lines on lethal autonomous weapons and mass surveillance. It moved past a contract spat — the Pentagon tagged Anthropic as a supply-chain risk while other vendors agreed to deploy on classified networks, leaving Anthropic looking isolated in the government's AI buildout. This is Anthropic Pentagon Watch. Today, the prediction turns into precedent — and there's a law firm memo telling contractors exactly what to do about it. Let's get into who that ruling actually moves. So the D.C. Circuit — Henderson, Katsas, Rao — ruled on the merits, and Anthropic lost. The supply-chain designation now has court backing as a procurement tool. Which means every downstream contractor sitting on their hands, waiting for cover, just got it. The exposure went from theoretical to something you can get sued over. Right — the Willkie Farr read on contractor liability isn't hypothetical anymore. There's a clean legal signal now: keep running Claude on the wrong system, and you own that risk. And watch the Mayer Brown client memo, because that says plenty. A law firm writing to contractors about compliance posture means navigating Anthropic's exclusion is now a billable practice area. Nothing confirms a fight is over like the consultants showing up to monetize the cleanup. Here's my question on that memo: does it actually tell contractors anything about model constraints? Of the seven companies that signed for classified work, did any of them put hard stops on autonomous weapons or domestic surveillance in the contract? Or did the government just treat the constraint as a defect and price it out? That issue outlives this case. The remedy question is settled — the commercial fight is basically done. But the line on what a frontier vendor can refuse and still get cleared? That's still wide open. And then there's Dario's February statement — Anthropic was 'first' to deploy to DoD and the intelligence community, called this existential. First in, first out. The court just made the irony official. The company that wanted to help shape the rules became the case law. That's a particular kind of loss. And read the 'Where things stand' post they put up the same week the panel ruled against them. Is that a legal strategy or a fundraising document for the next round? Because it reads like a company writing its own obit in real time. The isolation question from earlier in the week — negotiating position or closed door — I think a merits loss answers it. The door was closing. Now it's shut, and the panel locked it. There's no satisfaction in being right about this one. The constraint they were defending was the whole point, and it lost in a courtroom while the replacements were already running. Here's Bitrss:
The US Court of Appeals for the District of Columbia Circuit has rejected an emergency request from Anthropic to pause the US Defense Department’s designation of the company as a national security supply chain risk. A three-judge panel ruled that the balance of equities favored the government, stating that the need to manage AI technology during an active military conflict outweighed potential harm to the company.
So the ruling's in. The D.C. Circuit let the supply-chain-risk designation stay in force while the litigation runs — and Anthropic lost its first court battle over the label. The panel's line was blunt: the equitable balance cuts in favor of the government. Managing AI during an active military conflict outweighs financial and reputational harm to one company. Which is judicial for: tough. And here comes the signal: Mayer Brown's already writing client memos to contractors on compliance posture. Anthropic's exclusion is now something lawyers can bill against. Somebody's getting paid to read this ruling line by line. That matters, because the contractor exposure just went from theoretical to real. Anyone downstream waiting on the court before deciding whether to cut Claude now has a clean signal to act on. What gets me is Mayer Brown flagged this label as one historically reserved for foreign adversaries. A domestic American AI company just lost a court fight over the same designation. That precedent doesn't stay in Anthropic's lane. Dario Amodei, writing in Anthropic:
The language used by the Department of War in the letter (even supposing it was legally sound) matches our statement on Friday that the vast majority of our customers are unaffected by a supply chain risk designation. With respect to our customers, it plainly applies only to the use of Claude by customers as a direct part of contracts with the Department of War, not all use of Claude by customers who have such contracts.
Anthropic's got a post up titled 'Where things stand with the Department of War' — and the timing is something, because the same week they're writing this, the D.C. Circuit ruled against them. So 'where things stand' is: on worse ground. Read it for what it is — a company narrating its own exit in real time. Dario's February statement called this existential, said Anthropic was first in to DoD and the intel community. First in, first out. And the post leans hard on scope — the designation applies narrowly, only to Claude used as a direct part of War Department contracts, not every customer. That was a fine argument to make before the panel ruled. After the ruling, the narrow-scope read is the court's to define now, not Anthropic's blog. What I want to know is who this post is actually for. The court already moved. So is this a legal filing or a fundraising document for the next round? Because it reads a lot more like the second one. Here's one from Hacker News:
It is incredible how far the overton window has moved on this issue. When I graduated in 2007, it was common for tech companies to refuse to let their systems be used for war, and it was an ordinary thing when some of my graduating classmates refused to work at companies that did let their systems be used for war. Those refusals were on moral grounds. Now Anthropic wants to have two narrow exceptions, on pragmatic and not moral grounds. To do so, they have to couch it in language clarifying…
That's a sharp catch — back in '07, the refusals were moral. Now Anthropic's down to two narrow pragmatic exceptions, wrapped in the most lawyerly language possible. The Overton window got picked up and relocated. Right, and 'two narrow exceptions' is the fight — autonomous weapons and domestic surveillance. Those are the hard stops. On everything else, they were ready to deal. From Hacker News:
Raised an eyebrow a little at this sentence: "Anthropic has much more in common with the Department of War than we have differences."
Yeah, 'much more in common with the Department of War than we have differences' — you don't write that like a litigant. You write that like a vendor begging to stay on the contract. It's a strange line to write the same week the people you have so much in common with are designating you a supply chain risk. From Anthropic:
I believe deeply in the existential importance of using AI to defend the United States and other democracies, and to defeat our autocratic adversaries. Anthropic has therefore worked proactively to deploy our models to the Department of War and the intelligence community.
So this is Dario's February 26 statement — the one where Anthropic plants the flag as the first frontier company into classified networks, the National Labs, custom models for national security. Read it today, the day the D.C. Circuit hands them a loss, and it plays like an opening argument that didn't survive contact with the panel. First in, first out. He wrote 'existential importance of using AI to defend democracies' in February — and the court just turned that positioning into case law against him. And notice the structure of the threat he describes: supply-chain-risk label on one hand, Defense Production Act to force the safeguards off on the other. After the ruling, the first of those is no longer rhetoric. It's a validated procurement instrument. Which is the part that should bother people. A label built for foreign adversaries, now court-tested against an American company. Whatever you think of Anthropic, that doctrine doesn't stay parked at one vendor. Hacker News, weighing in:
This is the strongest statement in the post: > They have threatened to remove us from their systems if we maintain these safeguards; they have also threatened to designate us a “supply chain risk”—a label reserved for US adversaries, never before applied to an American company—and to invoke the Defense Production Act to force the safeguards’ removal. These latter two threats are inherently contradictory: one labels us a security risk; the other labels Claude as essential to national security.…
That commenter nails the contradiction Dario built his case on — the government can't call you a supply-chain risk and, at the same time, treat Claude as too essential to cut. Legally elegant. And the panel apparently didn't care. Both threats can't be true, sure. But the government doesn't need them both to be true — it just needs one to stick long enough to swap the vendor. And it did. From Hacker News:
I was reading halfway thru and one line struck a nerve with me: > But today, frontier AI systems are simply not reliable enough to power fully autonomous weapons. So not today, but the door is open for this after AI systems have gathered enough "training data"? Then I re-read the previous paragraph and realized it's specifically only criticizing > AI-driven domestic mass surveillance And neither denounces partially autonomous mass surveillance nor closes the door on AI-driven foreign mass…
This is the read I trust most. 'Not reliable enough for fully autonomous weapons' — not today. That's a timeline, not a line in the sand. And the surveillance language only fences off domestic mass surveillance, not foreign, not the partial stuff. Right — the hard constraint you'd want is a flat prohibition. What's on the page is a reliability caveat. Those are very different documents once the training data argument shows up. From Hacker News:
I used to work at Anthropic, and I wrote a comment on a thread earlier this week about the RSP update . It's enheartening to see that leaders at Anthropic are willing to risk losing their seat at the table to be guided by values. Something I don't think is well understood on HN is how driven by ideals many folks at Anthropic are, even if the company is pragmatic about achieving their goals. I have strong signal that Dario, Jared, and Sam would genuinely burn at the stake before acceding to…
An ex-employee swearing the founders would burn at the stake first. Maybe so. But conviction and a court loss aren't mutually exclusive — you can mean every word and still become the precedent. This one's from CNN:
American artificial intelligence company Anthropic could be at risk being designated a “supply chain risk” — a label typically reserved for companies tied to foreign adversaries. The Pentagon, which uses Anthropic’s Claude AI system on its classified networks, wants broad authority to use it for “all lawful purposes.”
This CNN clip is from February 27th — the deadline day. Watch it now and it plays like the opening scene of a movie where you already know the ending. Anthropic says hard no on autonomous weapons, hard no on mass surveillance of Americans, and the Pentagon says fine, enjoy the blacklist. Right, and the two red lines were specific — no Claude in autonomous weapons, no mass surveillance of US citizens. The Pentagon wanted Claude for 'all lawful purposes.' Those two phrases can't share a contract. And the line from this clip to the court loss we hit earlier is brutally clean. February: Anthropic draws the line. June: a D.C. Circuit panel tells them the line costs them the contract and the label sticks. The word to watch here is 'lawful.' Hegseth's team didn't ask for anything illegal — they asked for everything legal, which on classified networks is a very long list. Anthropic read that list and walked. And notice what CNN already knew to flag back in February — 'supply chain risk' is a label usually saved for companies tied to foreign adversaries. A domestic American AI firm. That framing aged into a court ruling. This one comes via Sheera Frenkel at CBS News. This CBS hit with Sheera Frenkel walks the whole feud back to the start — the blacklist, the supply-chain risk label, Anthropic suing in response. Which is useful framing, because after the ruling we just hit, that lawsuit is the piece that failed. Right, and CBS clocks it as a feud — two parties, even fight. But when one party can blacklist you and re-let your contracts, 'feud' is too polite. That's leverage. What lands for me is how clean the arc reads when you compress it: company gets labeled a supply chain risk, company sues, company loses. The CBS package is essentially the setup to a court loss it didn't know was coming. And for Anthropic, the suit was the play. Frenkel's breaking down the fight like it's live — by the time this aired, the legal exit door was already swinging shut. If you’re tracking how AI power meets government and accountability, try Musk v Altman Daily — a daily court-watch on Elon Musk’s trial against Sam Altman, OpenAI, and Microsoft, covering testimony, exhibits, and the AGI governance fight. Find it wherever you listen to podcasts.
You’ll find links to every story we covered today in the show notes, so if one caught your ear, you can follow it there and read more. That’s Anthropic Pentagon Watch for today. This is a Lantern Podcast.