The Pentagon just told the courts, 'we'd like a do-over' — it's appealing the order to drop Anthropic's supply-chain risk label, which means it's still paying lawyers to defend a designation a judge already blocked. This is Anthropic Pentagon Watch, and Gizmodo says Hegseth is losing his war. The Axios story underneath it says the Pentagon is only considering cutting Anthropic off. That gap is the whole show today. We've also got NSPM-11 out of the White House and a sharp piece on who's actually writing the rules for military AI. Spoiler: not Congress. And for once we've got names on the demands. Let's start there. So the 'Hegseth is losing' framing is tempting. But an appeal means the department is still fighting to keep the label alive. It lost a round; it hasn't quit. Sure, but look at what the appeal is protecting. Tony Johnson's piece names the two guardrails Hegseth wants gone: mass domestic surveillance and fully autonomous weapons. That's what 'all lawful purposes' cashes out to. Right — and it's the first sourced list of the actual guardrails. It reframes what those six Anthropic engineers at Fort Meade are actually being asked to strip out of an offensive operation. And the leverage to make them do it? Threatening the $200 million contract. You don't threaten to cut someone off if you can already force them. That threat is the Pentagon admitting it can't. So now you've got two tracks running at once: the designation appeal in the Ninth Circuit, and whatever enforcement directives come out of NSPM-11. Today's the first day we can confirm both are live at the same time. Speaking of NSPM-11 — it's addressed to 'The Secretary of War.' That title died in 1947. Either someone's making a point about the Pentagon's mandate, or nobody proofread a national security memo. We flagged that Friday. It's Monday and the title's still there. So it's not a typo anybody felt like fixing. Here's what changed for me. A week ago, I was asking whether Anthropic could hold the line and keep the money. They're holding the line. The Pentagon's the one on appeal after getting slapped in court. And the quieter story may be the sharpest: Indian Strategic Studies is framing this as a separation-of-powers gap. The executive branch is setting the rules because Congress isn't. That's the kind of argument you see in an appellate brief. It makes the national security label look thin, too. Hard to call Anthropic a risk for refusing autonomous weapons when that's exactly the kind of constraint a legislature would write down if it bothered to show up. The procurement-retaliation story has grown into a fight over who gets to set the rules: one arm appealing, one arm threatening the contract, and a memo reshuffling AI authority across the executive branch. The fight moved up a level. The White House writes:
Artificial intelligence (AI) will be among the most transformative technologies to national security in the history of the United States. When adopted appropriately, AI can help protect our warfighters during peacetime and on the battlefield, enable precise operations that minimize harm to civilians, and ensure the United States continues to maintain technical overmatch against our adversaries and strategic competitors.
So we finally have NSPM-11 in front of us — dated June 5, addressed to fifteen recipients. State, Treasury, Energy, DNI, CIA, OMB, the FBI, the National Cyber Director. You're looking at the whole national security apparatus here, way beyond one agency. And the second name on that list? The Secretary of War. That title died in 1947, Sarah. Either somebody at the NSC is making a point, or nobody proofread the memo. Both versions are a story. Section 1 says previous administrations imposed 'undue bureaucracy.' Translation: the guardrails are the bureaucracy. That's the framing running through the whole document. Right — and notice what it doesn't do. Fifteen addressees, lofty purpose language about technical overmatch, and not one concrete directive I can point to yet. It's a restructuring signal with the follow-on guidance still missing. And that's what I actually care about: has Treasury or OMB or DNI been handed anything actionable under this, or is it just rearranging who holds the AI authority on paper? Three days later, I don't see the operational text. Because the operational pressure is happening off this page. The memo gives you the architecture; the contract and the appeal show you the leverage. Here's Dave Lawler, Maria Curi at Strategic Study India:
The Pentagon is pushing four leading AI labs to let the military use their tools for "all lawful purposes," even in the most sensitive areas of weapons development, intelligence collection, and battlefield operations. Anthropic has not agreed to those terms, and the Pentagon is getting fed up after months of difficult negotiations.
Finally — names on the table. Anthropic's two no-go zones: mass surveillance of Americans, and fully autonomous weapons. That's what 'all lawful purposes' was always trying to bulldoze. And read the official's complaint carefully — they're not denying those are the lines. They're saying there's 'considerable gray area' and it's 'unworkable' to negotiate use-case by use-case. Unworkable. Translation: they don't want Claude unexpectedly refusing to run a domestic surveillance task. Sarah, they're stating the demand out loud. And it's four labs, not just Anthropic — same 'all lawful purposes' ask across the board. Anthropic's the one that hasn't folded, which is exactly why it's the one staring down a contract termination. Right. The leverage move here is the tell. If the Pentagon could legally force the guardrails off, they would. Instead they're threatening to cut the check. You don't reach for the wallet when you've got the law. Here's Mayer Brown:
The Pentagon told the California Northern District Court that it will be bringing U.S. District Court Judge Rita Lin’s March 26 ruling -- to prohibit federal agencies from designating Anthropic a “supply chain risk” and cutting the company’s technology from all federal systems -- to the Ninth Circuit Court of Appeals for reconsideration.
Update on the supply-chain-risk fight: the Pentagon filed a notice of appeal today, taking Judge Lin's March 26 injunction to the Ninth Circuit. So the appeal we were treating as hypothetical last week is on a live docket now. And read the timing. There was a seven-day stay on Lin's order from day one — Emil Michael said on X it was granted specifically so they could appeal. The appeal was baked in before the ink dried. Michael is very careful here — he says the 41 USC 4713 designation is, quote, 'in full force and effect,' and only the 3252 injunction is on hold. He's splitting the statutory tracks on purpose. So they're still fighting in court to keep the blacklist alive. A supply-chain-risk tag doesn't just touch the Pentagon, it can lock a company out of the entire federal government. That's the hammer they're appealing to keep swinging. From Tony Johnson at Indian Strategic Studies:
On Tuesday, Secretary of Defense Pete Hegseth gave Anthropic CEO Dario Amodei a Friday deadline: remove the guardrails on the Claude AI system — specifically, the company’s restrictions on mass domestic surveillance and fully autonomous weapons — or lose a $200 million Pentagon contract. Hegseth also threatened to invoke the Defense Production Act and designate Anthropic a supply chain risk, a designation that could effectively blacklist the company from work across the federal government.
Tony Johnson's framing in Indian Strategic Studies is the quiet one today, and maybe the sharpest: the people writing the rules for military AI right now are not in Congress. They're in a contract dispute over a $200 million deal. And Johnson gives the Pentagon its due — if taxpayers buy a capability, the military can argue it should deploy it for any lawful purpose. His point is that saying 'lawful' only starts the analysis; it doesn't end it. Here's what I've been waiting all week to put on the record: Johnson names the two guardrails. Mass domestic surveillance. Fully autonomous weapons. That's the 'all lawful purposes' Hegseth keeps waving around — strip those two, specifically. So when somebody says this is an abstract safety squabble — no. Hegseth handed Amodei a Friday deadline to drop restrictions on surveilling Americans and on weapons that pick their own targets. Put it in those words and the fight gets very simple. And the separation-of-powers angle is the appellate hook. If Congress never set these limits, an agency leaning on a contract to remove them is exactly the kind of authority question a court ends up refereeing. Gizmodo, with Matt Novak:
News broke this week that the National Security Agency (NSA) is using Anthropic’s Mythos AI model for offensive cyber operations, likely against China and Iran. It makes a lot of sense, given the reported power of Mythos to find and exploit vulnerabilities. But it’s just the latest sign that Defense Secretary Pete Hegseth might be losing his war against Anthropic.
So the headline is 'Hegseth Is Losing His War' — and the thing that makes it true is the NSA is already running Anthropic's Mythos for offensive cyber ops. The same company he labeled a supply-chain risk in March. Right. You don't blacklist a vendor whose engineers are sitting at Fort Meade customizing the model you're shooting at China with. Or you do, and you look like this. Novak's asking the right question: does the government need Anthropic more than it'll admit? Half a dozen engineers embedded at NSA say yes. You don't embed people in a supply-chain risk. I'd stress-test the verb, though. 'Losing' a round is real — the Pentagon's appealing the injunction, per Mayer Brown, and weighing a contract cutoff. A department doing all that is still fighting. Sure, but threatening to sever a $200M contract over the exact two guardrails — autonomous weapons and mass surveillance — because you can't force them to drop them? That's a tantrum with a budget line. If you’re tracking AI power and accountability, try Musk v Altman Daily — 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.
We’ve put links to all of today’s stories in the show notes, so if one of them deserves a closer read, you can find it there. That’s Anthropic Pentagon Watch for today. This is a Lantern Podcast.