Making sense of First Amendment free-speech claims sometimes first involves sorting out complex procedural and statutory pieces in a moving puzzle. As the White Russian-addled protagonist of The Big Lebowski would say, there might be “a very complicated case . . . a lot of ins, a lot of outs, a lot of what-have-yous.”

That’s certainly true when it comes to grasping strands and nuances of Anthropic’s First Amendment retaliation case against the US Department of War (DoW) and its secretary, Peter B. Hegseth. The artificial intelligence company’s claim hinges on the defendants allegedly punishing it for publicly expressing—and sticking to—its safety, risk, and civil liberties-grounded viewpoint that the DoW not deploy Anthropic’s Claude Gov model for “mass domestic surveillance of Americans and fully autonomous weapons.”

The procedural starting point for understanding Anthropic’s substantive legal theory is recognizing that it involves two lawsuits—one in San Francisco before US District Judge Rita F. Lin and the other before a three-judge panel of the US Court of Appeals for the District of Columbia Circuit. Each court recently delivered an important yet preliminary ruling: Lin’s March 26 decision favored Anthropic, while the appellate court’s April 8 order denied Anthropic’s motion for emergency relief but granted its request to fast-track proceedings.

Complicating matters is the fact that the stigmatizing, potentially business-destroying label of national security supply-chain risk slapped on Anthropic by Hegseth arises under two statutes. One law—10 U.S.C. § 3252—is front-and-center in Lin’s courtroom, while the other—41 U.S.C. § 4713, part of the Federal Acquisition Supply Chain Security Act of 2018—is center stage before the DC appellate court. That partly explains the different rulings. Here’s an overview of what’s transpired as it affects Anthropic’s First Amendment claim.

The March 26 Decision. Judge Lin issued a preliminary injunction against the government, concluding that “broad punitive measures” taken by the DoW against Anthropic “were likely unlawful” as “classic illegal First Amendment retaliation.” Those measures include designating Anthropic a national security supply-chain risk, banning it “from ever having another government contract,” and “announc[ing] that anyone who wants to do business with the US military must sever any commercial relationship with Anthropic.” That, Lin wrote, would mean “a company that used Claude to power its customer service chatbot could not serve as a defense contractor.”

The sanctions were imposed when contract negotiations between Anthropic and the DoW, which had used Claude without expressing any problems, collapsed in February. The breakdown came after Hegseth demanded that the DoW be allowed to deploy Anthropic’s AI tools for “all lawful uses.” Anthropic, however, refused to permit their deployment for mass surveillance and lethal autonomous warfare.

I previously described three key elements Anthropic must prove to win its First Amendment retaliation claim. Briefly, the company must demonstrate that its exercise of constitutionally protected expression was a substantial or motivating factor that caused it to suffer the material, adverse consequences meted out by Hegseth.

Lin’s opinion cites two points that significantly bolster Anthropic’s case: (1) AI safety is a matter of “important public debate,” and “Anthropic went public with its disagreement with” Hegseth about its safety concerns, including “criticizing the government’s contracting position in the press,” and (2) social media posts by Hegseth and President Donald Trump tie “Anthropic’s punishment to its attitude and rhetoric in the press,” with the two men making “repeated references to rhetoric and ideology.” For example, Trump posted on Truth Social that Anthropic is “A RADICAL LEFT, WOKE COMPANY” populated by “Leftwing nut jobs.”

In short, rather than leaving well enough alone, Hegseth and Trump are (at least at the preliminary injunction stage) being hoisted on their own petard. They could have simply terminated Anthropic’s contract and moved on; as Lin wrote, “everyone, including Anthropic, agrees that the [DoW] may permissibly stop using Claude and look for a new AI vendor.” Instead, they seemingly embraced Trump’s retributive proclivities, made Anthropic pay a steep price for publicly disagreeing with the DoW, and exposed—in Lin’s view—their ideological animus (the over-the-top, MAGA-chumming social media posts they apparently couldn’t resist making).

The April 8 Order. Things didn’t go quite as well for Anthropic on the other side of the country in early April. The DC appellate court denied Anthropic’s “emergency motion” for an “extraordinary remedy” to stay (delay) Hegseth’s determination under 41 U.S.C. § 4713 that the company “presents a supply-chain risk to national security” until the court fully reviews the case’s merits. The damning designation remains in place for now; the military cannot be forced “to prolong its dealings with an unwanted vendor of critical AI services in the middle of a significant ongoing military conflict.” The appellate court, however, expedited review of the case, and oral arguments are set for May 19 on the merits of Anthropic’s claims.