Anthropic Sues Pentagon Over AI Ethics Blacklist

TL;DR:

Anthropic filed two federal lawsuits on March 9 challenging the Pentagon's "supply chain risk" designation: the first time this label has been used against a U.S. company over a policy dispute.^[1]^[2]
The constitutional argument: Punishing a company for publicly stating it won't build surveillance tools violates the First Amendment.^[3]
Microsoft filed an amicus brief urging a temporary restraining order.^[4]^[5] So did 37 researchers from OpenAI and Google, plus 22 retired military officers.^[6]
Preliminary hearing set for March 24 in San Francisco federal court
The bigger picture: Can the government force AI companies to participate in mass surveillance by threatening their business?^[7]^[8]

Two Courts, One Message

On March 9, Anthropic's lawyers filed in two courtrooms simultaneously.^[1]^[9]

The first: a 48-page complaint in U.S. District Court for the Northern District of California, calling the Pentagon's actions "unprecedented and unlawful."^[3] This one names the Defense Department, Secretary Pete Hegseth, and over a dozen federal agencies as defendants.

The second: a narrower challenge at the U.S. Court of Appeals for the D.C. Circuit, targeting the supply chain risk designation specifically. Federal law gives that court special jurisdiction over these determinations.

Both lawsuits make the same core claim: the government punished Anthropic for expressing its views about how AI should and shouldn't be used. That's retaliation against protected speech.^[3]

The Constitutional Arguments

Anthropic's complaint lays out five counts. The First Amendment claims are the headline grabbers:

First Amendment retaliation: Anthropic says the Pentagon's designation punishes the company for "express[ing] its views, both publicly and to the government, about the limitations of its own AI services." When CEO Dario Amodei said Claude wouldn't be used for mass domestic surveillance, that was protected speech. The government's response, blacklisting the company, was punishment for that speech.

The lawsuit quotes Supreme Court precedent: the government "may not employ 'the power of the State to punish or suppress'" disfavored expression.

Fifth Amendment due process: The company also argues it never got a fair hearing. The supply chain risk designation carries severe penalties: hundreds of millions in lost contracts, reputational damage, a business-killing ultimatum to partners. All of that happened without Anthropic getting a meaningful chance to defend itself.

Administrative Procedure Act violations: The Defense Department allegedly failed to follow the rules Congress set for making these determinations. The supply chain risk statute exists to address foreign adversaries compromising U.S. systems, not to settle contract disputes with American companies.

Microsoft Jumps In

On March 10, Microsoft filed an amicus brief that reads like a warning shot.^[4]

The brief urges the court to issue a temporary restraining order blocking the designation while the case proceeds. Microsoft's argument: let the parties negotiate, like the Pentagon did with OpenAI.^[4]

But the brief goes further. Microsoft warns that the Pentagon's action sets a "dangerous precedent" by weaponizing a statute designed for foreign adversaries against a domestic company having a policy disagreement.^[5]

A Microsoft spokesperson put it plainly: "Everyone wants to ensure AI not used for mass domestic surveillance or to start a war without human control."^[4]

Microsoft has skin in the game. The company embeds Anthropic's technology in multiple products. Enforcing the ban would force "expensive product dismantling" and "substantial reengineering and legal costs." But the brief frames this as an industry-wide problem: any government contractor could face the same retaliation for holding the wrong views.^[5]

OpenAI and Google Employees Back Anthropic

The same day Microsoft filed, 37 researchers and engineers from OpenAI and Google DeepMind submitted their own amicus brief, in their personal capacities, supporting Anthropic.^[6]

These are Anthropic's competitors. They work for companies that stand to benefit from Anthropic's exclusion. And they're telling the court: this matters more than market share.^[6]

Microsoft's filing was also backed by 22 retired military officers, people who understand what AI tools could do in combat, and who apparently think some limits are worth preserving.

First Time for Everything

Here's what makes this case unusual: the supply chain risk designation has never been used this way before.

Microsoft's brief notes that the only previous use of this label against a company was Switzerland-based Acronis AG. That case involved foreign ownership concerns, the kind of threat the statute was written to address.

Anthropic is an American company. It passed security clearances for classified work. The Pentagon's own evaluation found Claude was "the most advanced and secure model for sensitive military applications."

Then Anthropic refused to change its policies, and suddenly it's a national security threat alongside Huawei and ZTE.

Legal analysts at Lawfare argue the designation "won't survive first contact with legal system." The statute requires evidence of foreign control or compromise. A CEO saying "we don't want our AI used for mass surveillance" doesn't fit.

The Privacy Question Nobody Elected

Here's the uncomfortable truth the EFF highlighted in their analysis: "the state of your privacy is being decided by contract negotiations between giant tech companies and the U.S. government."^[7]

Anthropic drew a line. OpenAI negotiated what critics call "weasel words." Neither outcome was decided by voters, legislators, or courts. It came down to what individual companies were willing to do.^[7]

The Center for American Progress is calling for congressional hearings. They want Amodei, Hegseth, and OpenAI's Sam Altman to testify under oath about what exactly the Pentagon asked for and what protections (if any) were actually agreed to.^[8]

Amodei himself said it best: "I actually do believe it is Congress's job" to address AI surveillance risks. Right now, that job is being done in contract negotiations and courtrooms.

What's Next

The preliminary hearing is set for March 24 in San Francisco federal court. Microsoft is pushing for a temporary restraining order to freeze the designation while the case proceeds.

If Anthropic wins, the precedent cuts deep: the government can't punish companies for their public statements about AI safety. That would be a significant check on the surveillance state's ability to pressure tech companies into compliance.

If Anthropic loses, the message is equally clear: hold the line on AI ethics, and you'll be treated like a foreign adversary.

Either way, this case will determine whether AI companies can maintain principles when the Pentagon comes calling, or whether refusing government demands makes you a national security risk by definition.

Sources

Published: March 16, 2026