Key Takeaways
- Anthropic has sued the U.S. Department of Defense after being labeled an “AI supply-chain risk”
- The designation effectively blocks defense contractors from using Claude
- The case puts AI safety policy and national security priorities on a collision course
- Legal experts say the outcome could shape how governments regulate dual-use AI for years
The First AI Supply-Chain War Has Moved Into a Courtroom
For years, the AI arms race was framed as a contest between nations. China versus the United States. State labs versus private ones. Speed versus safety. Nobody predicted it would end up here — with an American AI startup suing its own government.
On March 9, Anthropic filed two federal lawsuits against the Trump administration — one in the U.S. District Court for the Northern District of California, another in the D.C. Circuit Court of Appeals — after the Pentagon declared the company a national security supply-chain risk, requiring defense contractors to certify they don’t use Claude in any work tied to the Department of War.
The complaint calls the government’s actions “unprecedented and unlawful,” arguing the Constitution does not permit the government to punish a company for its protected speech on AI policy. For a startup whose enterprise business depends on the kind of institutional trust that a national security blacklist instantly erodes, the financial exposure is severe — with CFO Krishna Rao warning in a related filing that the government’s actions could reduce Anthropic’s 2026 revenue by multiple billions of dollars.
The full sequence of events that triggered the lawsuit is documented in the breakdown of how negotiations collapsed and OpenAI moved in.
Why the Pentagon Called an American Company a “Supply-Chain Risk”
The phrase usually appears in cybersecurity contexts — a tool historically reserved for foreign adversaries, not Silicon Valley startups. Applying it to a domestic AI company is, as Bloomberg noted in its coverage of the filing, a designation typically reserved for companies from countries the U.S. views as adversaries. It was the first time the federal government had used it against a U.S. company.
So what triggered it? Anthropic built its systems around a training approach known as Constitutional AI, designed to prevent models from generating harmful or unethical outputs. That safety architecture can block certain categories of prompts — including those involving surveillance or weaponization. For defense planners working on large-scale systems like the Pentagon’s Combined Joint All-Domain Command and Control initiative — a program designed to link AI across every branch of the U.S. military — restrictions like these aren’t philosophical disagreements. They are operational constraints.
One defense analyst put it plainly: “The Pentagon wants a Swiss Army knife. Anthropic is offering a blade that refuses to open if it senses a fight.”
The designation followed CEO Dario Amodei’s announcement that he would not allow Claude to be used for autonomous weapons or mass domestic surveillance — the same two red lines Anthropic had maintained throughout negotiations, and the same principles enshrined in its Responsible Scaling Policy.
The Procurement Rules Behind the Conflict
Anthropic’s lawsuit argues the designation violates federal procurement rules and constitutional protections on two separate grounds. The General Services Administration terminated Anthropic’s “OneGov” contract following a Trump Truth Social post directing every federal agency to immediately stop using the company’s technology — a move the lawsuit argues exceeds the authority Congress actually granted.
The key legal mechanism at the center of the dispute is Section 889 of the National Defense Authorization Act, which gives the U.S. government authority to block technologies deemed security risks within federal supply chains. Historically, those restrictions targeted foreign telecommunications and infrastructure providers. Applying a similar designation to a domestic AI company signals something new: AI models themselves are now considered strategic infrastructure.
Legal scholars are skeptical that the government’s case will hold. Writing in Lawfare, lawyers Michael Endrias and Alan Z.
Rozenshtein argued that the designation “exceeds what the statute authorizes.” He also noted that the required findings don’t hold up. Additionally, Hegseth’s own public statements may have undermined the government’s legal position even before litigation began.
Mayer Brown’s procurement team reached a similar conclusion. They flagged that the secondary boycott of Anthropic’s commercial clients appears to go beyond what the statute actually allows.
Timeline: How the Conflict Escalated
Early 2025 — Defense agencies begin evaluating generative AI tools for intelligence analysis and command systems.
Late 2025 — Anthropic pushes for strict contractual limitations on military use of its models.
February 21, 2026 — DoD lawyers introduce expanded usage rights. Anthropic rejects them.
February 24 — Amodei meets with Hegseth in a last-ditch attempt to reach a deal. The two fail to reach an agreement.
February 27 — The Trump administration orders federal agencies and military contractors to halt all business with Anthropic. Hours later, OpenAI signs its own Pentagon agreement — a pivot whose full implications are examined in the account of how the Pentagon contract was awarded.
March 4 — Anthropic receives formal written confirmation of the supply-chain risk designation under 10 USC 3252.
March 9 — Anthropic files two federal lawsuits, asking courts to vacate the designation, block its enforcement, and require federal agencies to withdraw directives to stop using the company’s technology.
The Global Context: How Other Powers Handle Military AI
The U.S. is not the only country wrestling with who controls AI in wartime — it is just the only one doing it in open court right now.
In China, private technology firms operate under regulations that give the government broad access to AI systems for national security purposes, with no equivalent right to refuse. The European Union has moved in the opposite direction, building risk classification frameworks under its AI Act that prioritize regulatory oversight over operational speed.
The U.S. sits in an uncomfortable middle — relying heavily on private labs for frontier AI capability while simultaneously trying to assert sovereign control over how that technology is deployed. That tension has now broken fully into the open.
The Procurement Ripple Effect
The Pentagon’s designation does more than sever federal contracts. Defense contractors, cloud providers, and intelligence partners routinely follow federal security guidelines when selecting vendors. Once a company carries a supply-chain risk label, partners often avoid it preemptively — not because they are required to, but because compliance audits are not worth the risk.
Anthropic has clarified that the designation applies narrowly — only to the use of Claude by customers as a direct part of contracts with the Department of War, not to all commercial use by companies that happen to have such contracts. But the chilling effect on enterprise partnerships is already visible, and it is playing out in real time across Washington’s defense contractor and developer ecosystem.
Notably, dozens of scientists and researchers at OpenAI and Google DeepMind — Anthropic’s two largest direct competitors — filed an amicus brief in their personal capacities supporting the lawsuit.
They argued that the supply-chain risk designation could harm U.S. competitiveness and hamper public debate about AI safety.
Rival lab employees publicly supported Anthropic’s position, showing that the industry widely views the designation as a threat to its ability to uphold ethical commitments.
Why This Lawsuit Could Become a Landmark
The outcome will draw a line that the entire industry has been circling for years. If the courts side with Anthropic, it establishes that AI companies retain the right to enforce ethical restrictions on their own technology — even when governments want broader access. If the government wins, it signals the opposite: that national security priorities ultimately override corporate AI safety frameworks.
Either outcome changes the rules permanently. For the first time in modern history, some of the world’s most consequential strategic technologies are being built not by governments, but by private labs. Those labs are now deciding, in courtrooms as much as in boardrooms, how far their creations can go.
And for the first time, a federal judge will have to answer that question.
