Anthropic vs. the Department of War

Anthropic, Palantir Technologies, and AWS announced a partnership to provide U.S. intelligence and defense agencies access to Claude 3 and 3.5 models. Claude would run inside Palantir's AI Platform (AIP), hosted on their Impact Level 6 accredited environment—one of the DoD's strictest security classifications, handling data up to "secret" level.

The partnership followed Anthropic's summer 2024 launch of Claude on AWS GovCloud and publication of its government acceptable use policy—contractual exceptions "carefully calibrated to enable beneficial uses by carefully selected government agencies." The exceptions permitted uses like legally authorized foreign intelligence analysis and combating human trafficking. But all standard restrictions remained in place: Anthropic's general Usage Policy continued to prohibit disinformation campaigns, the design or use of weapons, censorship, and malicious cyber operations. The government exceptions also did not extend to AI systems capable of domestic surveillance. These two categories—mass surveillance of americans and autonomous weapons—would become the fault line of the entire dispute.

Elizabeth Kelly, former head of the U.S. AI Safety Institute at NIST under Biden, joined Anthropic to lead a new "Beneficial Deployments" team. Anthropic also hired Tarun Chhabra (former Biden NSC official) for national security policy and Benjamin Merkel (former State Department appointee). The company confirmed hiring "more than two dozen people from all political backgrounds since the beginning of 2025."

To the Trump administration, this looked like a partisan hiring spree. Combined with Amodei's inauguration absence and Anthropic's lobbying against the Big Beautiful Bill's provision to preempt state AI regulation, the hires hardened the narrative that Anthropic was a "Resistance organization"—a label David Sacks would deploy months later.

The Pentagon's Chief Digital and AI Office announced $200M ceiling contracts with Anthropic, Google, and xAI, matching OpenAI's earlier award for a total potential of $800M across four companies. Anthropic's was a two-year agreement to "prototype frontier AI capabilities that advance U.S. national security." CDAO Chief Digital and AI Officer Doug Matty announced the awards.

This was the high-water mark of cooperation. Anthropic had secured a major defense contract while maintaining its usage restrictions. But the contract's terms would soon collide with the administration's ambitions.

Trump signed Executive Order 14347, "Restoring the United States Department of War," authorizing "Department of War" as a secondary title for the DoD and "Secretary of War" for Hegseth. The website defense.gov redirected to war.gov that same day.

The rebrand was more than symbolic. It signaled an institutional posture incompatible with the kind of restraint Anthropic's usage policies embodied.

On October 14, Sacks attacked Anthropic on X after co-founder Jack Clark published an essay about being "deeply afraid" of AI's trajectory, calling the company "principally responsible for the state regulatory frenzy that is damaging the startup ecosystem."

On October 20, he escalated further, calling Anthropic a "Resistance organization" and accusing it of trying to "backdoor Woke AI and other AI regulations through Blue states like California." He referenced Amodei's comparison of Trump to a "feudal warlord" during the 2024 campaign, his public support for Kamala Harris, and "the biggest Democrat mega-donors in tech."

Amodei responded on October 21 with a blog post pushing back against "inaccurate claims," stressing alignment with the administration on "key areas of AI policy" and noting he had met Trump at a July AI and energy summit in Pennsylvania.

U.S. special operations forces (JSOC) conducted overnight strikes across Caracas, Venezuela, capturing President Nicolás Maduro and his wife Cilia Flores. They were flown to New York to face federal charges including narco-terrorism and drug trafficking.

Claude was deployed during the active operation via Palantir's platform. Afterward, an Anthropic employee reportedly contacted a Palantir counterpart to inquire about how Claude had been used—a move that would later infuriate Pentagon officials, who interpreted it as Anthropic disapproving of its software's use in a kinetic operation. This inquiry became one of the proximate triggers for the February crisis.

Two major stories broke simultaneously on this Friday. The Wall Street Journal and Axios revealed Claude's use in the Maduro raid. A senior administration official told Axios the Pentagon would be "reevaluating its partnership" with Anthropic.

That same day, Anthropic announced Chris Liddell—former Deputy White House Chief of Staff under Trump's first term and ex-CFO of Microsoft—had joined its board, in what CNBC described as an attempt to "earn some goodwill."

On February 17, Semafor reported the details of the Palantir rift: during a routine post-raid check-in, an Anthropic official discussed the operation with a Palantir executive. The Palantir executive was "alarmed by the implication" that Anthropic might resist military use and reported the conversation to the Pentagon.

The Information published a leaked 1,600-word internal memo Amodei had sent to staff on February 27. He called OpenAI's Pentagon deal "maybe 20% real and 80% safety theater," called Altman's public statements "straight up lies" and "mendacious," and said the "real reasons DoW and the Trump admin do not like us" was that Anthropic hadn't "donated to Trump" and hadn't offered "dictator-style praise to Trump." He called OpenAI employees "sort of a gullible bunch."

The fallout was immediate. Treasury, HHS, State Department, GSA, Secret Service, NASA, Commerce, OPM, and FHFA all began unwinding Anthropic access. An FDA employee told NOTUS the decision "will basically wipe out 18 months of efforts." Altman acknowledged in his own internal memo: "We shouldn't have rushed. [...] it just looked opportunistic and sloppy."

By March 5–6, Amodei published a public apology: "It was a difficult day for the company, and I apologize for the tone of the post. It does not reflect my careful or considered views."

Three developments converge. Anthropic filed a motion for an emergency stay of the supply chain risk designation in the D.C. Circuit, citing estimated 2026 revenue losses of "hundreds of millions, or even multiple billions, of dollars."

At Palantir's AIPcon 9 in Maryland, CEO Alex Karp confirmed Claude remains in active use: "The Department of War is planning to phase out Anthropic; currently, it's not phased out. Our products are integrated with Anthropic, and in the future, it will probably be integrated with other large language models."

CBS News reported that Claude continues to support U.S. operations in the war with Iran, with the military "processing roughly a thousand potential targets a day" with AI-assisted intelligence analysis. The 180-day removal clock set by the Davies memo runs until September 2, 2026. The courts, the companies, and the administration are all racing against it.