What the AI ​​company says has happened

By Akash Sriram

March 9 (Reuters) – Anthropic sued the U.S. government on Monday, escalating a dispute over the AI ​​firm’s refusal to remove safety restrictions on its cloud model.

The Amazon-backed company has said it is willing to work with the military. Just not under any circumstances.

It also filed a related lawsuit in the U.S. Court of Appeals for the D.C. Circuit challenging the separate statutory authority the government sought.

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

What does anthropic say the conflict is about?

Anthropic said it has spent years building the cloud into the government’s most widely deployed frontier AI model, including classified military networks, creating a specialized version of “cloud government” and reducing many of its standard limitations for national security work.

The dispute began during negotiations over the Pentagon’s GenAI.mil platform in the fall of 2025, when the Department of Defense asked Anthropic to drop its usage policy entirely and allow the cloud to be used for what the government said was “all lawful uses.”

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

The company says Cloud has not been tested for this use and cannot do so safely. It said it had also offered to help transfer the work to another provider if no agreement could be reached.

Pentagon officials have offered a different account of how the conflict began. The department’s chief technology officer said publicly that tensions escalated after the U.S. attack on Venezuela, when an Anthropic executive called a colleague at Palantir to ask if the cloud had been used in the operation.

That account does not appear in Anthropic’s complaint.

From ultimatums to blanket bans

Defense Secretary Pete Hegseth met with Anthropic CEO Dario Amodei on February 24, delivering an ultimatum: comply within four days or face two penalties – mandatory under the Defense Production Act, or cut from the defense supply chain as a “national security risk”.

Amudi publicly denied the request on February 26. The next day, just before 5:01 a.m. Eastern time, President Donald Trump issued a directive on social media ordering every federal agency to immediately stop using Anthropic’s technology.

In a social media post, the president characterized Anthropic as a “radical left, whack company.”

Hours later, Hegseth announced at X that Entropic was a “supply chain risk to national security” and that no military contractor or supplier could do business with the company.

Agencies quickly fell in line. The General Services Administration has canceled Anthropic’s government-wide contract. Treasury, state, and federal housing finance agencies have publicly cut ties. Anthropic’s complaint alleges that the Pentagon launched a massive airstrike on Iran using Anthropic devices hours after the ban.

White House spokeswoman Liz Huston said the administration would not allow a company to “jeopardize our national security by how the largest and most powerful military in the world operates,” adding that the U.S. military would “never be held hostage by the ideological ambitions of big tech leaders” and would uphold the Constitution, saying “AI is of no service.”

Why did Anthropic decide to trial?

Entropy argues that supply chain design has no realism. The company points to its FedRAMP authorization, active security clearances, and years of government praise, including from Hegseth, who called the cloud’s capabilities “fantastic” at a February 24 meeting.

Two senior Pentagon officials later told reporters that there was “no evidence of a supply chain threat” and that the designation was “ideologically motivated.”

Anthropic raises five legal claims, arguing that the actions violate the Administrative Procedure Act, the First Amendment, the Fifth Amendment, the President’s statutory authority, and the APA’s unauthorized agency sanctions.

(Reporting by Akash Sriram in Bangalore)