Friday night, Anthropic killed two of its own flagship models — and by Saturday the White House wanted to talk. This is Anthropic Pentagon Watch. Today: Mythos and Fable go dark, two courts point in the same ugly direction, and a meeting nobody's calling a surrender. And somebody used the word 'productive.' We'll find out what that euphemism is hiding. Start with the shutdown. So, be precise about the instrument here — the trigger was Commerce's export-control directive, separate from the supply-chain designation and the presidential freeze. It forced Anthropic to pull Mythos and Fable that same night. All week we asked what happens to projects already running this stuff. The answer stopped being theoretical. The models are gone. And those were the newest flagships. You don't shut down your best products overnight if you're indifferent to the leverage — that question's answered too. Now, the courts. Inside Defense reads a recent procedural notice as a bad signal for Anthropic. Pair that with the DC Circuit denying the stay — per Jones Walker — and the legal picture gets ugly in two places at once. They're getting funneled to a table they didn't pick, on a clock they don't run. And that changes the negotiating math. The stay denial also means the designation is more likely to survive challenge — the condition the Pentagon needs before it writes acquisition guidance with real teeth. Now, the meeting itself. The Times calls it 'productive' — diplomatic language for: we're not telling you what got conceded. But here's the tell — the Times says officials believe Mythos could be 'critical for security.' That's the government signaling it wants to negotiate around this thing, not just seize it. Right, and who's brokering the compromise? The same executive branch that issued the designation and the Commerce directive. There is no clean regulator in that room. Which makes Amodei's public push for binding federal AI rules an interesting card to be holding when you walk in asking for your access back. So the guardrail question is back on the table, but now Anthropic's bargaining from a weaker legal position. The control language has the least outside protection right when it's most available to trade away. And there's a fresh waiver fight under all this. Defense contractors who lost Mythos access — who do they call to get it back, and what does that process cost? CFR flagged this scenario a week ago — military AI rules getting written through a contract dispute while Congress watches. Now there's a specific directive and a specific meeting to point at. Watch the compromise terms. 'Productive' is a word that survives until somebody publishes what changed hands. From Mayer Brown:
The Trump administration is following through with its threat to designate artificial intelligence company Anthropic as a supply chain risk in an unprecedented move that could force other government contractors to stop using the AI chatbot Claude.
The Pentagon's word for it is 'effective immediately' — and per Mayer Brown, the supply chain risk label is something the government has historically reserved for foreign firms tied to U.S. adversaries. They just hung it on a domestic American AI company. 'Effective immediately,' plus the line that the decision 'appeared to shut down the opportunity for further negotiation.' So the designation lands Thursday — and we'll get to the White House meeting later — but the official posture on Thursday was: no talks. Right, and the legal weight here is that this label can force other contractors to drop Claude. The fight isn't just over Anthropic's own contracts — it's everyone downstream who built on Claude. That's the part that actually bites. You don't have to cancel a single Anthropic deal to choke the company — just make every defense contractor scared to touch Claude. The label gets you what the lawsuit couldn't. Inside Defense, with Mariam Baksh:
Oral arguments on the merits of Anthropic's opposition to the Pentagon supply-chain risk designation will be heard by the same panel of judges that rejected the plaintiff's request for an emergency stay, according to a notice from the U.S. Court of Appeals for the District of Columbia Circuit, which court watchers say doesn't bode well for the company.
Here's the procedural tell. The same three judges who denied Anthropic's emergency stay back on April 8 — Henderson, Katsas, Rao — are now the panel hearing the merits on May 19. You don't get a fresh look. The judges who already told you no on the stay are the ones reading your full case. Court watcher Charlie Bullock put it gently — 'not a great development.' Precision matters here: denying a stay only means Anthropic hasn't shown it's likely to win. It doesn't decide the merits. But the same panel has already heard this pitch once, and it didn't move them. Right, and combine that with the designation we just hit — supply chain risk, effective immediately. The legal channel's narrowing, and there's only one exit door left. It happens to lead to the White House. This one's from Jones Walker LLP:
The government’s actions against Anthropic came in three forms: a “Presidential Directive” ordering every federal agency to permanently ban Anthropic, a “Hegseth Directive” barring military contractors from commercial dealings with Anthropic, and two supply-chain-risk designation letters signed the same day.
So here's the split-screen Jones Walker lays out. Thirteen days ago, Judge Rita Lin in California called the government's theory against Anthropic 'Orwellian.' Same dispute, parallel designation, and the DC Circuit panel? It denied the stay, set oral argument for May 19, and handed the parties three pointed questions to brief. Two courts, and only one of them used the word Orwellian. Guess which one Anthropic's lawyers are not quoting in their press release. Jones Walker's read is that the two rulings aren't actually contradicting — they're answering different questions about different instruments. But the practical math is brutal: a stay denial means the supply-chain designation stays live while they litigate. Which is the whole ballgame. A district judge calling the theory creepy doesn't help you if the appeals court won't pause the thing that's actually cutting off your contracts. Sympathy isn't an injunction. And put that next to the Inside Defense court-watcher note from earlier — the procedural signal looks bad. Now you've got two venues moving the same way as oral argument approaches. Julian E. Barnes; Sheera Frenkel; Tyler Pager, writing in The New York Times:
Anthropic’s chief executive met with White House officials on Friday for discussions that both sides described as “productive,” as the Trump administration works to forge a compromise that would bring the artificial intelligence company’s technology back into the government, according to U.S. officials and others briefed on the matter.
Friday at the White House: Anthropic's CEO and administration officials, both sides calling it 'productive.' In diplomatic English, that means nobody's saying who gave up what. 'Productive.' Put that word on a spike. The same week a court signals they're losing and Commerce orders their newest models dark — and now suddenly everyone's productive. The detail that matters: officials told the Times they believe Mythos could be 'critical for security.' That's the government looking for a bargaining path around the model instead of a straight seizure. Critical for security — and they switched it off anyway. So Anthropic walks into that room having just killed its own flagship. Sarah, the funnel ends at this meeting; it's the only exit left. And with the DC Circuit stay denied and Inside Defense reading the court notice as unfavorable, they're walking in off a two-court hit. The negotiating math collapsed before anyone sat down. Ars Technica writes:
Anthropic completely shut off access to its Mythos 5 and Fable 5 models Friday night, just days after they were launched. The move comes after Anthropic's receipt of a US Commerce Department directive Friday evening, subjecting the new models to export controls restricting their use anywhere outside the United States.
Here's the answer to the question we kept asking all week — what happens to live products under this fight. Friday night, Anthropic switched off Mythos 5 and Fable 5 for every customer on the planet. And the instrument that did it wasn't the Pentagon supply-chain label from the segment we just hit — it was a Commerce Department export-control directive. Different statute, sharper teeth, landed on a Friday evening. Seventy-two hours. Fable 5 was public for three days before Commerce killed it worldwide — not foreign users, everybody. That's the directive reaching places the blacklist couldn't. And read the reason out loud: a reported jailbreak that walked right around the classifiers on cyber, chem, and bio prompts. So the safety stack Anthropic sells as the differentiator — somebody broke it in 72 hours, and the government found out before the customers did. The Axios source says the pause buys the national-security apparatus a few weeks to harden. So this looks temporary by design — a short leash while the hardening happens. If you're following Anthropic's government and AI governance moves, you may also like Musk v Altman Daily — a daily court-watch on Elon Musk's trial against Sam Altman, OpenAI, and Microsoft. Find it wherever you listen to podcasts.
You'll find links to every story we covered today in the show notes. If one caught your attention, they're there for a closer read.
That's Anthropic Pentagon Watch for today. This is a Lantern Podcast.