A judge called Pentagon actions 'troubling' from the bench yesterday — and by the end of the day, eight AI companies had signed classified-network deals with the same Pentagon. Anthropic's week in one sentence. This is Anthropic Pentagon Watch. Today: two judicial rebukes on the federal record, the procurement perimeter getting eight names wider, and a White House executive order dropping into a governance fight senators are already calling obsolete. The Pentagon was in court defending a blacklist while it was signing classified deals with eight competitors. That's the market Anthropic is getting boxed out of, right there. All right, let's get into it. Chiara Rossi, writing in Startmag:
On Tuesday, during a hearing before a Washington appeals court, the Trump administration defended its decision to classify Anthropic as a supply chain risk, while actively seeking to understand how to use its most powerful model, Mythos, to counter cyber threats, reports Axios.
So the hearing is on the record now: the same administration that classified Anthropic as a supply chain risk was also trying to figure out how to use Mythos — Anthropic's most powerful model — for cyberdefense. That came out Tuesday, and the judge called Pentagon actions 'troubling.' That's the second judicial rebuke on the federal record in a week, after 'spectacular overreach' last episode. That starts to matter if this keeps climbing. The Pentagon's position in open court is basically: you're a national security threat, and also, can we borrow your best model? Come on — that's not a legal argument, that's a shakedown with better stationery. And while Anthropic was in that courtroom, the Pentagon formalized classified-network deals with eight other AI companies. So Anthropic is now outside a group of eight — not just behind OpenAI, but shut out of the structure the Pentagon built around this week. Those eight companies did not have to sue the Pentagon to get access. So if Anthropic wins, it becomes the only lab that got the DoD to negotiate use limits in court instead of a conference room. This one's from Bradivarius:
When U.S. District Judge Rita Lin called the Pentagon’s treatment of Anthropic “troubling,” she wasn’t just stating the obvious—she was highlighting a pattern of overreach. In my opinion, the Trump administration’s move to blacklist Anthropic feels less about security and more about sending a message to the tech industry.
Judge Rita Lin put 'troubling' on the federal record yesterday to describe the Pentagon's treatment of Anthropic — and that gives the case its second judicial rebuke, after 'spectacular overreach' from the panel earlier. Different judges, same direction. And the same day Lin is calling it troubling from the bench, the Pentagon is finalizing classified-network deals with eight other AI companies. Eight. The club expands while they're in court defending the blacklist. That makes the isolation argument a lot sharper. Anthropic isn't outside one deal with one rival anymore — it's outside eight formalized agreements. And if those rebukes keep stacking up, the government's deference argument gets harder to sell. And if Anthropic wins, what does it actually get? Restored access with no binding use constraints? Or a court finding of retaliation that turns into enforceable terms without a signed contract? Because those eight companies didn't go to a courtroom to negotiate their limits. Anthropic did. This one's from Yahoo News Canada:
Eight leading artificial intelligence companies have reached deals to deploy their technology in classified Pentagon computer networks, the Defense Department said Friday. The announcement leaves Anthropic - one of the industry’s key players - increasingly isolated as it battles the Trump administration in court over being branded a national security risk by the Pentagon.
We've been tracking the OpenAI defense-access thread all week, and this widened it a lot. Eight companies now have classified-network deals with the Pentagon, announced Friday. Anthropic is not one of them. And Emil Michael, the Pentagon's own CTO, went on CNBC and basically pitched Anthropic's blacklisting as vendor diversification. 'Irresponsible to rely on one partner' — sure, if by that you mean, 'we found eight companies who didn't try to put limits on what we do with their AI.' The procurement picture is concrete now: Anthropic is in court arguing it was wrongly locked out, while the Pentagon just formalized a group of eight that doesn't include them. That's a competitive perimeter being drawn in real time. And if Anthropic wins in court and gets access restored, it wins back into a structure where seven or eight other labs already said yes without conditions. That's leverage in one direction only — and it isn't Anthropic's. Here's Karen Freifeld, Courtney Rozen and Jarrett Renshaw at Reuters:
WASHINGTON, May 20 (Reuters) - U.S. President Donald Trump is expected to sign an executive order on AI and cybersecurity as soon as Thursday, two sources familiar with the matter told Reuters, as pressure grows from parts of his political base to increase oversight of new AI models, such as Anthropic's Mythos.
Reuters is reporting that Trump is expected to sign — or may have already signed, depending on your timezone — an executive order on AI oversight as soon as Thursday. The headline move is a voluntary framework where developers hand models to the government 90 days before public release, plus early access for critical infrastructure like banks. Voluntary. That word is doing a lot of work. Steve Bannon and Amy Kremer are the named pressure sources pushing for mandatory government security tests — so that's the political base forcing this, not DoD, not the intelligence community. The White House is trying to thread a needle between MAGA activists who want a hard gate and tech CEOs they're trying to line up for a signing photo. And the timing matters. Anthropic is in oral arguments this week, a judge just put 'troubling' on the federal record describing Pentagon conduct, and the White House is building a voluntary pre-release review framework at the same time. Those are two governance tracks that aren't talking to each other, and Anthropic's lawsuit is sitting right in the gap. If Anthropic wins in court and gets access restored with no binding use constraints, and the executive order produces a voluntary handshake with no enforcement teeth, you end up with a company that negotiated harder in court than eight other labs did at the conference table — and still lands inside a framework that can't actually compel anything. Army Times, with Michael Peck:
As the Pentagon requests a massive budget increase to develop autonomous weapons, some senators worry that DoD policy isn’t keeping pace. DOD’s “policy architecture really has to scale with it,” Sen. Joni Ernst, R-Iowa, said Tuesday during a hearing of the Senate Subcommittee on Emerging Threats and Capabilities.
Senator Joni Ernst said it out loud Tuesday: Pentagon policy 'really has to scale with it' — meaning autonomous weapons — and 'this is where we probably lag behind.' That's not a think-tank white paper; that's on the Senate subcommittee record, the same week Anthropic is in court and the Pentagon just cut eight classified-network deals. And the budget number tells you how far behind we are — DAWG goes from $225 million to $55 billion in one request. You don't move that fast on policy because you're ahead of the problem. Meanwhile Directive 3000.09, last updated in 2023, is already looking obsolete. This is the external check that's been missing all week. Anthropic goes to court to enforce its own use limits, and senators are saying the Pentagon's governance framework can't keep pace. Two separate tracks, same gap, and neither one is talking to the other. So the thing that would enforce any AI lab's red lines inside Pentagon networks is an institution senators are publicly describing as behind the curve. If Anthropic wins in court and gets access restored, who's actually holding Directive 3000.09 in one hand and an Anthropic API key in the other? Ernst's answer is basically: probably nobody adequate yet. If you follow the governance fights behind frontier AI, 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.
You'll find links to every story we covered today in the show notes, so if one caught your attention, that's the place to keep reading.
That's Anthropic Pentagon Watch for this Thursday. This is a Lantern Podcast.