The Department of Defense designated Anthropic — the company behind Claude — a "supply chain risk" in early March 2026. That label had historically been reserved for foreign adversaries like Kaspersky, Huawei, and ZTE. Anthropic is the first American company to ever receive it.

The reason? Anthropic refused to let the Pentagon use Claude for fully autonomous weapons and mass domestic surveillance of Americans.

That's it. That's the whole dispute.

Back in July 2025, Anthropic signed a $200 million, two-year contract with the Pentagon to deploy Claude across classified military networks. The deal was negotiated under the Biden administration. Things went sideways when the DOD started pushing for broader access during deployment negotiations on their GenAI.mil platform in September.

By January 2026, after Claude was used via a Palantir partnership in a Venezuelan operation, the Pentagon wanted more. Specifically, they wanted "all lawful use cases" — no restrictions, no guardrails, no exceptions.

Anthropic drew two lines:

On February 24, Defense Secretary Pete Hegseth gave CEO Dario Amodei a deadline: drop the restrictions or lose the contract. Amodei refused.

Three days later, Hegseth posted on X declaring Anthropic a supply chain risk. Hours later, Trump posted on Truth Social ordering all federal agencies to "immediately cease" using Anthropic's technology, calling it a "Radical Left AI company." The official designation letter followed on March 3.

Here's the part that really gets me.

On the same day the supply chain risk designation took effect — March 4 — the U.S. military was actively using Claude to identify and prioritize targets in Iran. The Pentagon was simultaneously blacklisting Anthropic and depending on its technology for active combat operations. Claude was so deeply embedded in operational systems that removing it on short notice was impossible.

The government also argued Claude was so critical to national security that they floated invoking the Defense Production Act to force Anthropic's compliance. So which is it? Is Claude a supply chain risk or an irreplaceable military asset? Can't be both.

Within hours of Trump's directive, OpenAI announced its own Pentagon deal. Sam Altman said they'd deploy models on classified networks with "technical safeguards" — nominally the same restrictions Anthropic wanted (no autonomous weapons, no mass surveillance), but structured as internal technical controls rather than explicit contract terms.

The optics were terrible. The backlash was immediate. Altman himself later admitted the deal "looked opportunistic and sloppy." By March 2, OpenAI was already amending the contract to add explicit prohibitions on domestic surveillance. So OpenAI ended up with essentially the same guardrails Anthropic asked for — they just got rewarded instead of punished.

Meanwhile, hundreds of OpenAI and Google employees petitioned against military AI agreements.

Anthropic sued on March 9. The case split across two courts, producing conflicting rulings.

San Francisco (Judge Rita Lin, March 26): Granted Anthropic a preliminary injunction blocking the supply chain designation. The judge found clear evidence of First Amendment retaliation, pointing to Hegseth's public statements calling Anthropic "arrogant" and accusing them of "corporate virtue-signaling" and "defective altruism." The ruling called the designation punitive rather than security-driven.

D.C. Circuit (April 8): Denied Anthropic's emergency stay, keeping the ban in force on national security grounds. However, the court expedited the case and scheduled oral arguments for May 19.

Legal experts at Lawfare have argued the government's position is "close to untenable." The statutes being invoked (FASCSA and 10 U.S.C. § 3252) were written to address foreign adversaries infiltrating supply chains through backdoors and sabotage — not domestic companies over contract disagreements. The legislative history only references foreign threats. And the secondary boycott — barring all contractors from doing any business with Anthropic — likely exceeds what the statute actually authorizes.

The ban backfired in one spectacular way: the public loved it.

Claude hit #1 on the U.S. App Store within days. Anthropic reported onboarding over one million new users per day, breaking internal records continuously. Daily signups tripled. The company went from relatively unknown outside tech circles to a household name practically overnight.

It turns out that telling Americans a company was punished for refusing to build autonomous weapons is a pretty effective marketing campaign.

The D.C. Circuit oral arguments on May 19 will address three threshold questions, including whether the court even has jurisdiction over the case. The outcome could set major precedent:

This matters well beyond Anthropic. The case will define whether the US government can effectively punish American AI companies for maintaining safety restrictions. If the designation stands, every AI company bidding on government work will know the price of saying no. If it falls, there's a real check on using obscure procurement statutes as political weapons.

The irony is thick: the Pentagon punished the one company that insisted on guardrails, rewarded the one that didn't (at first), then the winner ended up adopting the same guardrails anyway. The technology the government called a "risk" was simultaneously being used in live combat operations.

Whatever your politics, the legal question is clean: can Cold War-era supply chain statutes designed to keep Huawei out of government networks be repurposed to blacklist American companies that won't remove safety features? May 19 might give us an answer.