A sworn declaration says Pentagon officials told Anthropic the two sides were nearly aligned — just one week after Trump called the whole thing dead. Here's how we got here: Anthropic's fight with the Pentagon began over restrictions on military use of Claude, then became a First Amendment retaliation case. Anthropic says the supply-chain-risk designation punished it for its public safety stance rather than any proven security flaw. A federal judge has already paused the government's action after finding that the cutoff looked retaliatory, and the case is still moving forward. This is Anthropic Pentagon Watch. Today — a courtroom contradiction, two lobbying filings that don't match, and Dario explaining why he won't sign a certain letter. Stick around. TechVersusMan writes:
Anthropic submitted two sworn declarations to a California federal court late Friday afternoon, pushing back on the Pentagon's assertion that the AI company poses an "unacceptable risk to national security" and arguing that the government's case relies on technical misunderstandings and claims that were never actually raised during the months of negotiations.
Two sworn declarations landed late Friday, and here's the line that matters: Anthropic says the Pentagon's whole 'unacceptable risk' case rests on claims that never came up during months of negotiations. Meaning: the Pentagon invented the grievance after the fact. Right. The new filings say Pentagon negotiators privately saw the two sides as nearly aligned. Under oath. And this was during the same window when Trump was out there calling the relationship dead. That's the part I can finally interrogate instead of speculate about. You can't tell a company in private that you're close to a deal, then tell a federal judge in public that it's an existential threat. One side of that contradiction is sworn testimony now. And a judge already paused this thing because it looked built to punish Anthropic's safety stance, not fix a proven flaw. So I want a name — who at DoD signed off on calling Anthropic an 'unacceptable risk' while their own negotiators were saying the opposite? The declarations answer the government point by point. That makes its appellate theory a lot harder to build. It's tough to construct a national-security emergency around a deal your own people said was nearly done. So does that internal contradiction actually matter in court? The Pentagon was privately signaling it was close to a deal with Anthropic just days after Trump publicly declared the relationship dead. Or is presidential rhetoric mostly noise in a fight over a supply-chain-risk designation? It could matter a lot. In its March 9th suit, Anthropic calls the designation an 'unlawful campaign of retaliation' for refusing to allow unrestricted military use of Claude. Its argument is that the government punished the company and dressed that punishment up as a national-security finding. And the timeline is hard to ignore. Lawfare reported that Trump directed every federal agency on Truth Social to stop using Anthropic's technology immediately. Hegseth issued the supply-chain-risk designation almost right afterward. Trump's post came first. On March 26th, the Northern District of California judge granted a preliminary injunction, blocking enforcement of the label while the case moves ahead. Defense One also reported that legal experts called the underlying thinking 'dubious' and ideologically motivated rather than grounded in a real risk assessment. If the Pentagon was telling Anthropic at the same time that they were nearly aligned on the contract terms, that directly undercuts the claim of a genuine, immediate supply-chain threat. It gives Anthropic real evidence that the designation was used as a pressure tactic. But the Pentagon is still formally maintaining the designation after the injunction. Does winning that preliminary round actually change anything on the ground for Anthropic? Not yet. Law.com reported the day after the injunction that the Pentagon was still deeming Anthropic a supply-chain risk. MIT Technology Review noted that the government had seven days to appeal, and a second case challenging the designation is still unresolved. The injunction freezes enforcement, but it doesn't erase the label. Watch whether the government appeals to the D.C. Circuit or the Ninth Circuit, and whether those private conversations about a near-deal surface in discovery. If they do, they become evidence, not just context. Michael Kan, writing in PCMag:
An open letter urging the US to support rather than restrict open-weight AI models has received public support from all the major tech companies—except Anthropic. CEO Dario Amodei is now explaining why, suggesting that open AI models could empower attackers more than defenders.
So Amodei finally put a name to the silence. Seventy-seven companies signed that open-weights letter — from Microsoft and Nvidia to OpenAI, Google, even SpaceX. Anthropic was the one holdout. Now he's explaining why: open models hand more to attackers than defenders. And he's careful about it, too. He says he agrees with much of the letter and has never advocated a ban. The position comes with a lot of hedges. Here's what gets me: on the same day, he'll make a public, principled case about open weights, but only when somebody else's model is on the table. Meanwhile, his lawyers are two beats away from arguing that the Pentagon misread Anthropic's own safety stance. That's the asymmetry, yeah. In court, it's 'you got our technology wrong.' On the blog, it's 'here's my technology opinion about everyone else's release.' Same safety vocabulary, two very different jobs. I just want him to put both ideas in one sentence: restrictions for competitors, an exemption for Anthropic. Then call it defender advantage. At least now it's on the record instead of just an empty chair on a signatory list. Next up: whether the reasoning survives contact with the export-control filings. DEV Community writes:
Anthropic's second-quarter federal lobbying disclosure names export controls and model distillation repeatedly and by statute. OpenAI's, filed for the same period, does not mention China, open weights, export controls, or distillation at all. The contrast lands one day after the New York Times reported, on anonymous sources, that both companies privately pressed Washington to restrict Chinese open-weight models - a claim neither filing confirms and neither company has denied on the record.
Here's the receipt the Times couldn't give you. Anthropic's Q2 lobbying disclosure names export controls and distillation by statute, over and over — the AI Diffusion Rule and the AI OVERWATCH Act. OpenAI's filing for the same quarter? No mention of China or open weights, and nothing on distillation. Zero. And these are LD-2 filings on the Senate's own disclosure system — sworn, public, and dated July 20. One day after the Times ran anonymous sources saying both labs privately pushed Washington on Chinese models. One company named the issues on the record. The other didn't. The caveat matters, though. The form records subject matter, not position. A company arguing to loosen export controls can file the exact same line as one arguing to tighten them. So this tells you what Anthropic paid attention to, not what it asked for. Fine, but attention costs money: $1.97 million for Anthropic, compared with OpenAI's $1.2 million that quarter. You don't name distillation seventeen ways across the House and Senate, Commerce and Treasury, for fun. That's the same posture their lawyers are running in the California case, just with a dollar figure attached. And that brings us back to the filing we opened with. In court, Anthropic's a First Amendment plaintiff; in the disclosures filed that same month, it's lobbying on export controls. Both are documented now. You don't have to pick which Anthropic is real. The paperwork shows you both. For more on the infrastructure behind AI, check out The Data Center Daily. It covers hyperscaler spending, the power grid, semiconductor supply, and the energy markets being reshaped by intelligence at scale. Find it wherever you listen to podcasts.
You'll find links to every story in today's show notes if you want to spend more time with anything that caught your attention.
That's Anthropic Pentagon Watch for today. This is a Lantern Podcast.