Product2 distinct publishers3 min readPublished
Rita Lin vacated the supply-chain risk label and the order keeping military suppliers away from Anthropic, which turns model-vendor political risk into an exposure a buyer can actually bound. The Pentagon can still buy elsewhere.
The Product Desk · Product desk
Compiled by The Product DeskSomething wrong?How this is made
The evidence that decided this was the government's own paperwork. In March, when Lin first blocked the designation, she quoted Defense Department records showing Anthropic had been named a supply chain risk because of its "hostile manner through the press," which she called classic illegal First Amendment retaliation, according to The Verge [12]. Thursday's 59-page ruling adds that the government is still discussing collaboration with Anthropic on a new model, Mythos, in sensitive contexts, which Lin read as inconsistent with a genuine fear that the company would poison its own software [11][3]. The Pentagon's stated reason had been that giving Anthropic access to classified systems would introduce unacceptable risk because the lab could disable or alter its technology at a time of war [9], and its own file did not support that.
Timing is the part worth copying into a spreadsheet. The designation was dated February 27 [1]. Anthropic sued in a California district court in March, and Lin temporarily blocked the blacklist that same month [13][12], which puts an upper bound of about 32 days on how long the ban ran before a court touched it [18]. Getting to a merits ruling took until Thursday, the parallel case at the US Court of Appeals for the District of Columbia is still open [10], and Wired reports the department is expected to appeal [6]. The Verge described the fight as monthslong [17], which is the right unit for this kind of exposure.
What survived the ruling is the buyer's discretion. Lin wrote that the Department of War is "undisputedly free to select the AI vendor of its choice" even as she found the measures against Anthropic illegal and baseless [3], and she affirmed that the Pentagon does not have to use Anthropic's models at all [5]. Winning removes the penalties and leaves the $200 million Claude deal where it fell [7][5]. Seven other labs, among them Google, Microsoft, OpenAI and SpaceX, signed the any-lawful-use terms Hegseth wanted [14]. Anthropic kept its two limits, on lethal autonomous weapons and on mass surveillance [8], and paid for them in litigation.
For a vendor risk register, two axes do most of the work. The first is whether the exposure runs through a buyer's discretion or through a coercive instrument that reaches into your own stack, the way the supplier ban did [2]. The second is whether you can swap the vendor inside one change window. Discretionary and swappable is ordinary churn, worth pricing as churn. Discretionary and sticky is the expensive quadrant, and no ruling helps there [5]. Coercive and swappable argues for doing the swap rather than tracking a docket. Coercive and sticky is the only quadrant where a vendor's appetite for litigation belongs in your scoring, and it is the quadrant this ruling repriced, because the instrument that reached past the buyer turned out to be reviewable and was reversed for nine agencies at once [2][4].
The bound Lin drew is narrow and real. An agency can decline to buy a model for reasons it never explains, and it can be made to undo a blacklist imposed because the vendor argued in public about terms [3][12].
Ranked by verification strength, evidence, and original report placement.
US district judge Rita Lin in California vacated defense secretary Pete Hegseth's February 27 decision labelling Anthropic a "supply-chain risk" and making the company ineligible for federal contracts, ruling the designation amounted to unconstitutional retaliation.
The ruling also lifted an additional punitive measure by Hegseth that prevented contractors or suppliers to the US military from doing business with Anthropic, a position Lin called "arbitrary, capricious, an abuse of discretion, and otherwise not in accordance with law."
In a 59-page ruling Lin wrote: "Though the Department of War is undisputedly free to select the AI vendor of its choice, the evidence demonstrates that the broad measures imposed on Anthropic were illegal and baseless."
Lin ruled that nine agencies, including the Pentagon, the Treasury Department, the State Department and the Department of Homeland Security, had improperly imposed sanctions on Anthropic, and the ruling removes those penalties.
Lin affirmed that the Pentagon is not required to use Anthropic's models, and the ruling does not prevent the Pentagon from electing to use other models.
The case stems from a dispute between the Pentagon and Anthropic over a $200 million deal to use the lab's Claude models for military applications.
Distinct publishers with included, body-backed reporting in this cluster.
1 article · August 27, 2026
2 articles · August 27, 2026
Follow any of these and your For You feed starts watching them — no settings page required.
build
Anthropic's federal ban falls on a record that failed the supply-chain statute1 distinct publisher
product
Washington's secret AI test is coming for open weights, and release dates go with it2 distinct publishers
invest
The 81% Problem: AI's Star CEOs Are Polling Badly With The People They Need To Hire1 distinct publisher
product
Nine AI leaders, nine majorities of distrust: the floor onboarding copy cannot lift1 distinct publisher
Evidence-backed comparisons of source perspectives and observed adoption signals. Read the methodology
Which Builder, Operator, and Investor concerns the observed source mix emphasized—not a truth score.
Evidence, demonstrated adoption, hype gap, incentives, and confidence are assessed independently, each on its own current evidence. How these are measured.
The court's own words, twice over
Almost everything load-carrying here is a direct quotation from a 59-page opinion, and WIRED and The Verge quote the same sentences without straining them. What is missing is the document itself: neither outlet links or paginates the ruling, so a reader takes the quotes on trust. The one claim with no visible source is WIRED's expectation of an appeal.
Eligible again, nothing re-bought
The ruling restores what Anthropic lost on paper across nine agencies, but the only concrete procurement facts in our coverage run the other way: seven rival labs signed while Anthropic was frozen out, and the court explicitly declined to make the Pentagon a buyer. The one forward signal is the judge's note that talks over the Mythos model were live in sensitive contexts.
The word 'win' outruns the remedy
Both outlets reach for victory framing, The Verge most directly, and the legal finding earns it. The commercial reading does not. A district judge removed penalties and said the buyer may still buy elsewhere; a DC case is unresolved; an appeal is expected. The overstatement is modest and mostly lives in the headlines, not the bodies, because WIRED itself supplies the limits.
Only one side is talking
The vindicated party is the only party quoted. Anthropic's statement appears in both accounts; the Pentagon could not be reached and no official responds anywhere in our coverage. The Verge builds much of its chronology from Amodei's own public statement about democratic values, which is exactly the narrative the ruling now flatters. Set against that, the rival labs that took the contracts have their own reasons to stay quiet, and nobody asked them.
Solid at the core, soft at the edges
The holding is safe: two publishers, same quotations, same judge, same day. Confidence drops on the particulars only one of them carries, the nine agencies and the $200 million on WIRED's side, the seven replacement deals on The Verge's. And the two accounts cannot agree on the name of the Anthropic spokesperson delivering an otherwise identical quote, which is the sort of slip that argues for reading the single-sourced numbers with one eyebrow raised.