Appeals court backs the Pentagon's exclusion of Anthropic over Claude's usage limits

The contract that started this fight is small next to the money now in play. Anthropic's prototype agreement to put Claude on classified networks had a $200 million ceiling ^[4]. Investors have discussed a listing that could value the company near $2 trillion and raise as much as $100 billion, figures that are still preliminary and unconfirmed ^[23]. Measured against those numbers, the ceiling is 0.01% of the valuation and 0.2% of the raise ^[21][22]. The damage Anthropic claims is much larger. According to Reuters, the company says the designation has cost it billions of dollars in lost business ^[9], and executives' court filings also warned of delayed enterprise deals ^[15].

I think the contractor bar is the likeliest route from $200 million to billions. The designation keeps the military off Anthropic's models and also blocks defense contractors from using them in their work for the department ^[8]. Enterprise customers supply the large majority of Anthropic's revenue ^[24]. The bar has limits: it covers work that touches covered military systems ^[13]. The company has not disclosed how much of its revenue comes from customers who do that kind of work.

The government relied on two statutes, and the outcome divided along them. A judge in the Northern District of California struck down one designation under 10 U.S.C. 3252, a statute that limits supply-chain risk to malicious acts by adversaries ^[6]. Ars Technica reported that the appeals court did not dispute that finding. Instead it reviewed the second designation under 41 U.S.C. 4713, where, the majority wrote, "no such bad motive is required" ^[6]. Judge Gregory Katsas wrote that Trump and Defense Secretary Pete Hegseth "must determine how best to balance the competing risks" ^[7]. Judge Karen LeCraft Henderson, dissenting, said the statute was aimed at hostile actors infiltrating federal systems, not "a contractor's honest and upfront enforcement of restrictions on a covered article's use disfavored by the government" ^[16]. The panel also rejected Anthropic's argument that the label was retaliation for its public stance on AI safety ^[11].

From here the case can go three ways. If the ruling holds, the Pentagon and its contractors stay off Claude for covered work. If further review reverses it, the label comes off but the Pentagon still has no deal it wants: the department sought unrestricted access for all lawful purposes ^[14], and its officials argue that no private vendor should dictate how the military deploys technology on the battlefield ^[20]. In the third, Anthropic drops the two limits. It says certain uses cross ethical lines it will not cross ^[19]. "We respectfully disagree with the court's decision," an Anthropic spokesperson said ^[18].

In my view, any valuation model should put direct Pentagon revenue at zero, because the only outcome that restores it is the one the company has ruled out. The contractor exposure is still open. Calling it a settled risk is premature while the panel's order is on hold pending further review ^[17] and the broader contractor-wide ban stays blocked by the San Francisco court, which called it punitive and unsupported ^[10]. The zero is wrong if the department would accept a vendor's use limits after losing in court, and its officials' stated position says it would not ^[20].