A federal judge shields Anthropic's AI use-case limits from government retaliation

The order draws a line between two government powers that procurement teams tend to treat as one. Lin's injunction leaves the government free to award the work to another AI firm, or to buy nothing at all, so long as it acts consistently with "applicable regulations, statutes, and constitutional provisions" ^[5]. What it removes is the power to convert a contracting disagreement into a penalty, which the judge described as a desire to make a public example of the company for its "arrogance" in criticizing the government ^[6]. A vendor holding a use-case line can still lose the award, and what the injunction closes off is the other risk: being made an example for holding that line.

The costly part of the punishment traveled through third parties rather than through the contract. Government lawyers argued that the supply chain risk designation did not stop companies from using Anthropic's technology for other purposes ^[13], but Lin found that Hegseth's February post telling military contractors they could not conduct "any commercial activity with Anthropic" produced a broader boycott, reaching contractors that were using the company for work unrelated to the military ^[12]. For a vendor weighing whether to hold a use restriction, that contagion through the customer base, rather than the single lost award, is the real exposure, and it is the part now enjoined.

The theory the government lost on repays close reading by anyone drafting acceptable-use language. Officials argued that Anthropic's tough negotiating stance and its push for AI restrictions gave grounds to fear "future sabotage" through a "kill switch" in the software ^[9]. Lin called that concern entirely unfounded, relying on Anthropic's evidence that it cannot maintain backdoor access to its systems, and on the difficulty of squaring a sabotage fear with the government's continued interest in collaborating on advanced models ^[10]. The lesson for drafters is that a use restriction defended as a policy and contractual commitment holds up better when the vendor can also show it retains no technical ability to reach into a deployed system.

The opinion reaches three separate instruments ^[18]: the supply chain risk designation, whose legal definition Lin held Anthropic does not meet ^[8]; the February contractor post ^[12]; and the agency cutoffs that followed the president's Truth Social order to "IMMEDIATELY CEASE all use of Anthropic's technology," which she found were not authorized by law ^[14]. The sentence most likely to be quoted across a negotiating table is narrower than the constitutional holding: an IT vendor, Lin wrote, "does not become a potential adversary of the United States whenever it asks probing questions or stubbornly insists on particular contracting terms" ^[11].

The scope of this ruling is limited: one district judge, one set of defendants, an opinion that can still be appealed, so nobody's standard terms changed on Thursday. That much is accurate. The supplied record does not say whether the government will appeal; CBS News reported that it had approached the Pentagon and the White House for comment ^[16], and Anthropic said it welcomed the ruling that the designation was unlawful ^[15]. What exists in hand is a 59-page factual record ^[6] and a permanent bar on the specific measures ^[1]. For terms being drafted this quarter, that is enough to change who pays for the carve-out: a buyer who wants it removed has to bargain it away, because the coercive route is enjoined and the record now includes the sequence Lin noted, in which the punishment was broadcast on social media before the administrative findings for the designation had begun ^[17].