The Pentagon just filed its notice of appeal. So the government is done waiting out the court order and is now actively trying to keep Anthropic tagged as a supply-chain risk. This is Anthropic Pentagon Watch. We end the week with a government brief in the record and an appeal in motion — and in a second, we’re getting to the Pentagon CTO quote that puts this whole fight in a very uncomfortable light. And yes, we’re naming OpenAI today. Anthropic said no, OpenAI said yes, and that gap on autonomous weapons stops looking accidental pretty fast — it looks like a business advantage. This one's from Inside Defense:
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.
The Pentagon filed its notice of appeal today in the Northern District of California, sending Judge Rita Lin’s March 26 ruling — the one that blocked the supply-chain-risk designation and cut Anthropic’s tech off from federal systems — to the Ninth Circuit. Under Secretary Emil Michael had already signaled this on X the day Lin’s order dropped, so that seven-day stay was always a runway, not a reprieve. Emil Michael posted on X that the 41 USC 4713 designation is "in full force and effect" while the injunction is on hold. That’s the Under Secretary of Defense for Research and Engineering telling the internet a court order doesn’t apply to him yet. And now they’re spending appellate bandwidth to try to make that stick. All week we were asking whether the government would actually fight the lift. Well, there’s your answer — filed, docketed, and headed to the Ninth Circuit. The week started with a contested label and ends with the Pentagon pressing it hard. That’s a real posture change, not just paperwork. And remember what’s now on the record from Al Jazeera’s coverage of the filing: the government’s theory is that Anthropic refusing to strip out guardrails is what justified the designation. So the appeal isn’t just defending a procurement call — it’s defending the idea that a vendor’s use policy can get it blacklisted. OpenAI said yes to DOD terms; Anthropic said no. One of them is in the Ninth Circuit now. Al Jazeera writes:
The administration of United States President Donald Trump has said in a court filing that the Pentagon’s blacklisting of Anthropic was justified and lawful, opposing the artificial intelligence company’s high-stakes lawsuit challenging the decision.
The government finally put its theory in writing: the designation was justified, it wasn’t retaliation, and Anthropic’s First Amendment claim is unlikely to win. That’s not a talking point — that’s a filed brief, which means a federal judge has to deal with it on the record. Read the filing, though. The trigger was Anthropic refusing to "release the restrictions" on autonomous weapons and domestic surveillance. The government is telling a court that saying no to those use cases is what got you blacklisted. That’s the admission. And that loops right back to what we flagged earlier this week — DOD was actively running Claude in operational contexts while it was also building the legal case that Claude’s guardrails made Anthropic a supply-chain threat. Now that contradiction lives in a live brief, not just a leak. The First Amendment framing is the fight to watch. The government says this is a contract dispute, not retaliation for speech. But "you refused to remove restrictions" is a very specific description of speech they wanted Anthropic to remove. From David Mortlock, Britt Mosman, David Levine at Willkie Farr & Gallagher LLP:
The scope and immediate effect of Secretary Hegseth’s order are in dispute, but regardless of how the ban is implemented, the incident has highlighted the risks of more aggressive actions against federal contractors.
The Willkie Farr client alert from March 2 is worth pulling back up now that the Pentagon has filed its notice of appeal — because when the designation first landed, neither the Truth Social post nor Hegseth’s X directive cited any legal authority. Anthropic’s own statement was the first document to name 10 U.S.C. § 3252. The government’s lawyers are now defending a label that their own officials didn’t bother grounding in statute when they announced it. And now the Pentagon CTO is on the record calling Anthropic’s guardrails "not democratic." So the sequence is: no legal citation on day one, a court order pushing back, and then the ideological rewrite once the legal theory starts wobbling. That’s not a legal strategy; that’s a retreat dressed up as principle. The Al Jazeera filing coverage finally gives us the brief itself — and the explicit theory is that Anthropic’s refusal to remove guardrails justified the designation. Willkie flagged back in March that the scope and effect of Hegseth’s order were "in dispute." They were right; they just couldn’t know that dispute would end up in appellate court three months later with the government doubling down. Breaking Defense, with Sydney J. Freedberg Jr.:
"What we're not going to do is let any one company dictate a new set of policies above and beyond what Congress has passed," Under Secretary Emil Michael said of the ongoing impasse between the Pentagon and Anthropic.
Under Secretary Emil Michael told Breaking Defense it’s "not democratic" for Anthropic to limit military use of Claude — his actual claim is that a private company setting guardrails is overriding what Congress passed. That is a remarkable thing to say on the record, because the flip side is obvious: the executive branch can label that same company a supply-chain risk over a policy disagreement with no similar democratic check. Emil Michael just gave us the quiet part in writing. The Pentagon’s CTO is calling vendor use policies antidemocratic while the executive branch unilaterally brands a company a national security threat — no vote, no statute, no hearing. That inversion is doing a lot of work. And now that inversion is in appellate form — the Pentagon filed its notice of appeal this week, so they’re spending government legal resources defending a designation their own CTO is publicly reframing as a constitutional principle. We started the week asking whether the government would put its theory in writing. They did, and the theory is basically that Anthropic’s safety limits are the democratic threat. Meanwhile, OpenAI accepted DOD terms, Anthropic didn’t, and international law still has exactly zero binding definitions for autonomous weapons to close the gap those two choices opened. The companies that stayed quiet aren’t bystanders — they’re the direct beneficiaries of the loophole the compliant lab just locked in. Here's Firdevs Bulut Kartal at Anadolu Agency:
Anthropic CEO Dario Amodei recently said the company declined to accept the Pentagon’s proposed contract terms, citing insufficient safeguards related to its Claude AI model. He warned that removing certain protections could enable the model’s use in mass surveillance of Americans or its integration into fully autonomous weapons systems. The Defense Department rejected those assertions. Spokesperson Sean Parnell said the military has no intention of surveilling US citizens or developing autonomous weapons without human involvement.
Anadolu Agency names OpenAI directly — they accepted the DOD terms Anthropic refused, which means there’s now a public, named split between two frontier labs on the same policy question. That’s not an industry divide in the abstract; it’s a fork in the road with specific companies on each side. And Anadolu is saying the part nobody in Washington wants to say out loud: international law hasn’t defined autonomous weapons or mass-surveillance thresholds, so DOD gets to write its own definitions domestically — and OpenAI just agreed to operate inside whatever those definitions happen to be on a given Tuesday. The legal gap Anadolu is describing isn’t abstract — it’s the vacuum that makes the government’s appeal worth filing. If there’s no binding external standard, DOD’s shifting domestic definitions are the only game in town, and right now the only company contesting them is the one they already tagged as a supply-chain risk. OpenAI staying quiet all week while Anthropic took the hit — that silence now has a price tag: exclusive access to a defense market where your compliant competitor just got labeled dangerous. The loophole doesn’t help everyone equally; it helps whoever signed first. Got feedback, story leads, or a correction for us? Send a note to anthropicpentagonwatch at lantern podcasts dot com. We read what comes in, and your tips help sharpen the show.
We’ve linked every story from today’s briefing in the show notes, so if there’s one you want to spend more time with, that’s the place to go next.
That’s Anthropic Pentagon Watch for this Friday, May 29th. Thanks for listening, and have a good weekend. This is a Lantern Podcast.