解读：Anthropic起诉政府案：这家AI公司称发生了什么

Explainer: Anthropic’s case against the government: what the AI company says happened

By Akash Sriram
March 9, 2026 8:27 PM UTC Updated 3 hours ago

节点运行失败
Item 1 of 2 Anthropic logo is seen in this illustration taken May 20, 2024. REUTERS/Dado Ruvic/Illustration/File Photo

[1/2]Anthropic logo is seen in this illustration taken May 20, 2024. REUTERS/Dado Ruvic/Illustration/File Photo

• Companies

• Anthropic PBC Follow
• Palantir Technologies Inc Follow

March 9 – Anthropic sued the U.S. government on Monday, escalating a dispute the AI company frames as retaliation for refusing to remove safety limits on its Claude model.

The Amazon-backed company said it was willing to work with the military. Just not on any terms.

Jumpstart your morning with the latest legal news delivered straight to your inbox from The Daily Docket newsletter. Sign up here.

It has also ​filed a related case in the U.S. Court of Appeals for the D.C. Circuit challenging a separate legal authority the government ‌invoked.

Advertisement · Scroll to continue

The following account is based on allegations made by Anthropic in its lawsuit.

WHAT ANTHROPIC SAYS THE DISPUTE IS ABOUT

Anthropic said it spent years building Claude into the government’s most widely deployed frontier AI model, including on classified military networks, developing a specialized “Claude Gov” version and loosening many of its standard restrictions to accommodate national security work.

The conflict began ​in the fall of 2025 during negotiations over the Pentagon’s GenAI.mil platform, when the Department of Defense demanded Anthropic abandon its usage ​policy entirely and allow Claude to be used for, in the government’s words, “all lawful uses”.

Advertisement · Scroll to continue

Anthropic said it largely agreed, ⁠except on two points it considered non-negotiable: it would not allow Claude to be used for lethal autonomous warfare without human oversight or for mass ​surveillance of Americans.

The company says Claude has not been tested for those uses and cannot perform them safely. It said it also offered to help ​transition the work to another provider if no agreement could be reached.

Pentagon officials have offered a different account of how the dispute began. The department’s chief technology officer said publicly that tensions escalated after a U.S. raid in Venezuela, when an Anthropic executive called a counterpart at Palantir to ask whether Claude had been used in the operation.

That ​account does not appear in Anthropic’s complaint.

FROM ULTIMATUM TO ALL-OUT BAN

Secretary of Defense Pete Hegseth met with Anthropic CEO Dario Amodei on February 24, ​presenting an ultimatum: comply within four days or face one of two punishments – compulsion under the Defense Production Act, or expulsion from the defense supply chain as a ‘national ‌security risk’.

Amodei ⁠rejected the demand publicly on February 26. The next day, before a 5:01 p.m. Eastern deadline had expired, President Donald Trump posted a directive on Truth Social ordering every federal agency to immediately cease all use of Anthropic’s technology.

In the social media post, the president characterized Anthropic as a “RADICAL LEFT, WOKE COMPANY”.

Hours later, Hegseth announced on X that Anthropic was a “Supply-Chain Risk to National Security” and that no military contractor or supplier could do commercial business ​with the company.

Agencies fell in line quickly. ​The General Services Administration terminated ⁠Anthropic’s government-wide contract. Treasury, State, and the Federal Housing Finance Agency publicly cut ties. The Anthropic complaint alleges the Pentagon launched a major air attack on Iran using Anthropic’s tools hours after the ban.

White House spokeswoman Liz ​Huston said the administration would not allow a company to “jeopardize our national security by dictating how the greatest ​and most powerful military ⁠in the world operates,” adding that U.S. forces would “never be held hostage by the ideological whims of Big Tech leaders” and would follow the Constitution, “not any woke AI company’s terms of service”.

WHY ANTHROPIC DECIDED TO SUE

Anthropic argues the supply chain designation has no factual basis. The company points to its FedRAMP authorization, ⁠active security ​clearances, and years of government praise, including from Hegseth, who called Claude’s capabilities “exquisite” at the ​February 24 meeting.

Two senior Pentagon officials subsequently told reporters there was “no evidence of supply-chain risk” and that the designation was “ideologically driven”.

Anthropic raises five legal claims, arguing the actions violated the Administrative ​Procedure Act, the First Amendment, the Fifth Amendment, the president’s statutory authority, and the APA’s prohibition on unauthorized agency sanctions.

Reporting by Akash Sriram in Bengaluru

Our Standards: The Thomson Reuters Trust Principles.