Anthropic sued the Pentagon this week, and the complaint says the government wanted two very specific things: fully autonomous weapons without human targeting oversight, and mass domestic surveillance. This is Anthropic Pentagon Watch. Today we’ve got a blacklist turning into a lawsuit, a seven-company classified-deployment list that makes the money side very explicit, and a federal open-source workaround that seems to undercut the government’s own story. And Dario Amodei calls the Pentagon the "Department of War" in the filing — which, no, that is not a stray phrase. That’s in the complaint, so we should probably start there before we get cute about anything else. Yeah, we’ll get there. A lot changed this week, and now both sides are in active litigation at once, which is the kind of detail you don’t bury three paragraphs down. From AI Weekly:
The Pentagon's supply chain risk designation against Anthropic is the first such label applied to a US company rather than a foreign adversary, creating a documented precedent that national security classifications can be used as leverage against domestic AI firms refusing specific use cases. Anthropic held two firm red lines: fully autonomous weapons without human targeting oversight, and mass domestic surveillance of US citizens.
Here’s where the blacklist fight stands: the California injunction is still in place, the D.C. Circuit turned down the government’s separate stay request, and Anthropic answered by suing. So now it’s live litigation on two fronts, not just one ongoing dispute. And those two red lines are spelled out in the federal complaint now: fully autonomous weapons without human targeting oversight, and mass domestic surveillance of U.S. citizens. So if the designation was punishment for that, a court has to look at it head-on. Dario Amodei using "Department of War" in a legal filing is not a typo. It’s a choice, in writing, under signature, and that’s exactly the kind of thing a Ninth Circuit panel is going to notice when it reads the First Amendment claim. Meanwhile, Axios says the government was still using Claude for Iran-related work while the blacklist was active, and Anthropic is now folding that into its retaliation argument. So they banned the vendor and kept the product. That’s the whole week, basically. Here's VJOURNAL:
The Defense Department said on May 1 that seven AI companies can now bring systems into classified networks, formalizing a shortlist for lawful operational use inside military workflows. AP reported that the new agreements cover Amazon Web Services, Google, Microsoft, Nvidia, OpenAI, Reflection, and SpaceX, and that at least one agreement included language requiring human oversight for autonomous or semiautonomous missions.
The May 1 roster is now explicit: Amazon Web Services, Google, Microsoft, Nvidia, OpenAI, Reflection, and SpaceX — seven companies cleared for classified network deployment. Anthropic is not on that list. That’s the procurement wall, right there. And the AP detail is doing a lot of work here: at least one of those agreements already includes human oversight language for autonomous missions. That’s Anthropic’s red line. So the companies that got in accepted the guardrail, and Anthropic got blacklisted for insisting on it. The word benchmark matters here. Classified-network access isn’t just another contract tier anymore — it’s the line between companies operating inside military AI and companies suing to stay adjacent to it. Anthropic filed suit this week, and everyone else on this roster either took the terms or never had to say no. OpenAI is inside that wall. SpaceX is inside that wall. Reflection — which a lot of people couldn’t pick out of a lineup — is inside that wall. So whatever the sorting mechanism was, it wasn’t capability. It was compliance posture. From Esther Shittu at AI Business:
Just before a deadline set by the Pentagon asking for Anthropic to relax some of its safety guardrails, President Donald Trump announced the government would stop working with the AI model provider. "I am directing EVERY Federal Agency in the United States Government to IMMEDIATELY CEASE all use of Anthropic's technology," President Trump said in a post on Truth Social.
So the AI Business piece nails down the timeline: Trump’s Truth Social post landed just before a Pentagon deadline for Anthropic to relax its guardrails, and then Hegseth’s "Department of War" supply-chain designation followed. For the lawsuit filed this week, that sequence is not decorative — it’s the causal chain Anthropic wants the Ninth Circuit to see. Hegseth said, "this decision is final," in a social post, and now Anthropic has a federal complaint with two named red lines: fully autonomous weapons without human targeting oversight, and mass domestic surveillance. That’s what "final" is colliding with. The six-month phase-out window Trump announced is also where the supply-chain designation’s compliance clock came from, and now that clock is being fought on two litigation tracks at once. What started as a Truth Social post is now a federal lawsuit and a pending Ninth Circuit appeal. And per Cedar Key Beacon, federal staff are apparently routing around that phase-out with open-weight Claude derivatives anyway. So the government banned the vendor, is fighting to keep the ban alive, and is still using the underlying work. That’s the week in one sentence. This one's from Cedar Key Beacon:
The conflict started when Anthropic refused to allow its AI, Claude, to be included into fully autonomous weaponry systems or utilized for widespread domestic monitoring. The corporation treated certain ethical constraints as non-negotiable architecture rather than discretionary choices, including them directly into its models. The Pentagon encountered an obstacle in its pursuit of “all lawful use” and unrestrained access.
Cedar Key Beacon ran a piece dated May 31 about federal staff in Washington reading Anthropic briefing materials for a model they’re officially barred from using. And that matters because the designation hit the vendor relationship — it never reached the model weights themselves, and weights can be downloaded. So the government blacklisted the vendor and is quietly using the vendor’s underlying work anyway. That’s not a loophole — that’s the procurement label admitting, on its own terms, that this was never really about model risk. And that quietly answers the "first domestic firm" framing we’ve been tracking all week. If agencies can route around the ban with open-weight derivatives, then the designation was always a vendor problem, not a technology problem. The label said supply-chain risk; the behavior says something else. Military News, with Konstantin Toropin:
TAMPA, Fla. (AP) — The Trump administration is pushing to unleash the power of artificial intelligence for the U.S. military while facing calls to put up guardrails around the rapidly developing technology from some companies — and even notes of caution from top leaders in uniform. Adm. Frank Bradley, head of U.S. Special Operations Command, told attendees of a recent annual special forces conference in Tampa, Florida, that troops “have to be very careful about how we come to (AI’s) employment and its inspiration into the delivery of lethality.”
Admiral Frank Bradley, who heads Special Operations Command, told a Tampa conference last week that troops have to be "very careful" about AI’s role in lethal strikes — and that humans need confidence the system will, quote, "deliver violence only where we intend it to be delivered." That’s a SOCOM four-star, on the record, drawing almost exactly the same line Anthropic put in its complaint. The legal point is real: when a federal court looks at Anthropic’s First Amendment claim, Bradley’s public remarks become ambient evidence that the constraint Anthropic refused to waive isn’t some fringe objection. It’s the stated view of the officer commanding U.S. special operations. If Anthropic Pentagon Watch helps you keep up with the story, subscribe and leave a quick review wherever you’re listening. It’s a small thing, but it helps other people find the show.
You’ll find links to every story we mentioned today in the show notes. If one caught your eye, that’s the place to dig in a little more.
That’s Anthropic Pentagon Watch for today. This is a Lantern Podcast.