Anthropic Sues Pentagon: Claims, Revenue Exposure and Two Cases

Two courts, one day

On Monday, March 9, 2026, Anthropic filed two separate lawsuits over the same designation: one in federal district court in California, one in the U.S. Court of Appeals for the D.C. Circuit. Both argue that the supply chain risk label the Department of War had issued four days earlier was retaliation for protected speech rather than a genuine security finding. [1]

The filing argues that "the Constitution does not allow the government to wield its enormous power to punish a company for its protected speech" and that "no federal statute authorizes the actions taken here." It describes the lawsuit itself as a last resort: "Anthropic turns to the judiciary as a last resort to vindicate its rights and halt the Executive's unlawful campaign of retaliation." [1]

Anthropic's public statement struck a different register than the filing, framing the suit as compatible with its government work rather than opposed to it: "seeking judicial review does not change our longstanding commitment to harnessing AI to protect our national security, but this is a necessary step to protect our business, our customers, and our partners." [1]

The dispute the designation grew out of

The designation traces back to two conditions Anthropic would not drop from its government terms of service: no use of Claude for mass domestic surveillance, and no use in fully autonomous weapons. CEO Dario Amodei made that refusal public on February 26, and Defense Secretary Pete Hegseth's department wanted a vendor that would accept any lawful government use of its models instead, without those carve-outs. [1]

Atlas interpretation: President Trump's later order to phase Claude out of federal agencies over six months was justified by citing classified military systems, including operations involving Iran, that were said to depend on it. That framing sits awkwardly next to the underlying dispute: the government was arguing Claude was embedded deeply enough in sensitive military work to require an unwind, from a company it was simultaneously trying to punish for refusing to let Claude be used more broadly in military and surveillance work. [1]

What Anthropic said was at stake

Anthropic was projecting roughly 14 billion dollars in 2026 revenue at a company valuation of 380 billion dollars, with more than 500 customers paying at least a million dollars a year for Claude access. The designation required every contractor with a Claude integration anywhere in its stack, defense related or not, to certify it had removed the models or lose its own Pentagon work. [1]

Atlas interpretation: The dollar figures in the filing are a measure of leverage, not just of harm. A company with 14 billion dollars of revenue mostly unrelated to defense work has far more to lose from a certification requirement that reaches every customer's stack than from losing Pentagon business directly, which is what made the designation an effective lever regardless of whether Anthropic's own defense revenue was large. [1]