Security1 distinct publisher2 min readPublished
The Pentagon labeled a vendor a supply chain risk by social media post, then conceded in court it had no evidence for the kill-switch capability it cited. Procurement should read designations as claims, not findings.
The Watch · Security desk

Compiled by The WatchSomething wrong?How this is made
The mechanics matter more than the politics for anyone who has to act on a label like this. Designating a supplier a supply chain risk requires the defense secretary to complete a specific set of actions, and Judge Rita Lin found that Hegseth did not complete them [8]. Letters to congressional committees asserted that less drastic steps had been evaluated and deemed not possible, and supplied no further detail [9]. The opinion runs 43 pages working through that gap [2].
The usage history cuts against reading the label as a product finding. The government ran Anthropic's Claude through much of 2025 without complaint, according to court documents [15]. Defense employees reached it through Palantir and had to accept a government-specific usage policy that cofounder Jared Kaplan said prohibited mass surveillance of Americans and lethal autonomous warfare, though his declaration to the court did not include the policy's terms, so that description is his and not a public document [16]. The disagreements started only when the government moved to contract with Anthropic directly [17].
Dates, because the adjectives in this fight are carrying a lot of weight. Trump's post was February 27 [6]; the order came March 26 [21], which is 27 days of federal agencies operating under an instruction to stop using a named vendor [23]. The seven-day appeal clock from the order puts a government filing at roughly April 2 [22]. The order is temporary [1], the government is expected to appeal [19], and Anthropic's second case in DC, built on a different segment of the law, has not been decided [4]. The company remains persona non grata with the government while that runs [5].
On the constitutional side, Lin wrote that the government "set out to publicly punish Anthropic for its 'ideology' and 'rhetoric,' as well as its 'arrogance'" and found the company on solid ground in its First Amendment claim [13]. She read the designation as effectively naming Anthropic a saboteur of the government, and found insufficient evidence for that [14]. Her order also forbids the government from fulfilling what Hegseth and Trump promised [24]. Dean Ball, who worked on AI policy in the Trump administration and filed a brief supporting Anthropic, called it a devastating ruling in which Anthropic was likely to prevail on essentially all of its theories [18]; that is a supporter's characterization of a preliminary order, not a final judgment.
The practical residue for a vendor review: a designation whose supporting record cannot be produced on demand is one party's position in a contract dispute, and it should be weighted that way alongside your own testing.
Ranked by verification strength, evidence, and original report placement.
A California judge temporarily blocked the Pentagon from labeling Anthropic a supply chain risk and from ordering government agencies to stop using its AI; the order was issued last Thursday.
Anthropic has a second case against the designation, filed in DC, that has yet to be decided and that makes similar allegations while referencing a different segment of the law.
Until the matter is settled, Anthropic remains persona non grata with the government.
President Trump's Truth Social post on February 27 referenced "Leftwing nutjobs" at Anthropic and directed every federal agency to stop using the company's AI.
Distinct publishers with included, body-backed reporting in this cluster.
Follow any of these and your For You feed starts watching them — no settings page required.
product
A court vacated the Defense Department's supply-chain-risk label on Anthropic1 distinct publisher
build
Anthropic's federal ban falls on a record that failed the supply-chain statute1 distinct publisher
product
Court reversal turns the Pentagon's Anthropic ban into a migration teams may have to run twice1 distinct publisher
product
Judge rules the Pentagon cannot punish Anthropic for keeping its use limits1 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.
Judicial record, single relay
The load here rests on a 43-page opinion and the filings behind it, including admissions the government's lawyers made on the record — the kill switch it could not evidence, the contractor ban it conceded had no force. Facts of that provenance rarely wobble. What keeps the number off the top of the scale is that MIT Technology Review is the only outlet carrying it in our coverage, the opinion is quoted in fragments rather than at length, and the Pentagon speaks only through its litigation positions.
Paperwork visible, consequences uncounted
We can date every official act — the February 27 directive, the March 3 filing, the March 26 order — and we know Claude was in Defense hands through much of 2025 via Palantir. What nobody has counted is the part that matters operationally: which agencies actually pulled the tool, what contracts lapsed, whether the Palantir channel kept running, and how contractors reacted to a ban that turned out to be unenforceable. The scaffolding is documented; the usage underneath it is not.
Verdict language ahead of a preliminary order
"Backfired" and "devastating" describe a temporary injunction. To its credit, MIT Technology Review flags the seven-day appeal window and the undecided DC case in its first paragraph, and the harshest verdict in the piece belongs to Dean Ball, who filed a brief on Anthropic's side. The record firmly supports 'the government lost round one for want of a record'; it does not yet support an ending, and the piece's own closing note — that contractors will avoid Anthropic anyway — quietly undercuts the win.
Every speaker has a side, and shows it
Read the roster: the president and the Defense Secretary posting first and lawyering after; Anthropic's cofounder describing a usage policy he did not attach; a former Trump AI-policy hand who wrote a brief for Anthropic supplying the quotable verdict; a company that had spent 2025 selling to defense while marketing itself on safety. Motives are dense here — but the reporting names each affiliation as it goes, and the government's contrary interests are visible through its own concessions rather than hidden. The pressure to distort exists; the disclosure mostly holds.
Firm on the ruling, open on the outcome
What the judge did on March 26, and why, is about as solid as reporting on a live case gets. Everything forward-looking is soft: an appeal that is expected rather than filed, a DC case whose theory is unexplained, and no way to size the commercial damage on either side. Add that we are reading a late-March account well after its appeal clock ran, and the safe conclusion is narrow — the designation failed for lack of a record, and nothing in our coverage says what replaced it.