Anthropic was the only major AI company to refuse the Pentagon's demand that it remove safety guardrails from its Claude AI model for military use — specifically, prohibitions on mass domestic surveillance and fully autonomous weapons. The standoff, which escalated from late 2025 through early 2026, culminated in the Pentagon designating Anthropic a "supply chain risk," President Trump ordering all federal agencies to stop using Anthropic's technology, and a federal judge blocking those actions as unconstitutional retaliation. Every other major AI lab — OpenAI, Google, and xAI — accepted the Pentagon's terms. The episode became a defining test of whether AI companies would maintain safety commitments under government pressure.
The relationship began on November 7, 2024, when Anthropic, Palantir Technologies, and Amazon Web Services announced a three-way partnership to bring Claude to U.S. intelligence and defense agencies. Claude was integrated into Palantir's AI Platform (AIP) and hosted on AWS infrastructure with Impact Level 6 accreditation, enabling processing of classified data up to the Secret level. Palantir CTO Shyam Sankar stated at the time: "Palantir is proud to be the first industry partner to bring Claude models to classified environments." Claude became the first frontier AI model deployed on the Pentagon's classified networks.
On July 14, 2025, the DoD's Chief Digital and Artificial Intelligence Office (CDAO) awarded Anthropic a two-year prototype Other Transaction Agreement with a $200 million ceiling to "prototype frontier AI capabilities that advance U.S. national security." The CDAO simultaneously awarded identical $200 million contracts to OpenAI, Google, and xAI — roughly $800 million total. Critically, the Pentagon agreed at signing to abide by Anthropic's acceptable use policy (AUP), which prohibited two specific applications: mass domestic surveillance of Americans, and fully autonomous weapons systems operating without meaningful human oversight. The scope of work included intelligence analysis, operational planning, modeling and simulation, and cyber operations.
The relationship fractured after a January 2026 incident in which Claude was reportedly used during the U.S. military operation that seized Venezuelan President Nicolás Maduro. According to Semafor's reporting, during a routine check-in between Palantir and Anthropic, an Anthropic official discussed the operation with a Palantir executive, who gathered that Anthropic disapproved of its technology's involvement. That executive reported the conversation to the Pentagon. Anthropic called this account "false."
The confrontation centered on two specific safety provisions in Anthropic's acceptable use policy. Defense Secretary Pete Hegseth's January 9, 2026 AI strategy memorandum directed that all DoD AI contracts incorporate standard "any lawful use" language within 180 days — eliminating company-specific guardrails entirely. The memo stated that the Pentagon would only deploy models "free from ideological constraints that limit lawful military applications."
Anthropic drew two explicit red lines. On mass domestic surveillance, CEO Dario Amodei wrote in a February 26, 2026 official statement: "We support the use of AI for lawful foreign intelligence and counterintelligence missions. But using these systems for mass domestic surveillance is incompatible with democratic values." Anthropic argued that under current law, the government can purchase detailed records of Americans' movements and web browsing from commercial sources without a warrant, and that powerful AI makes it possible to "assemble this scattered, individually innocuous data into a comprehensive picture of any person's life — automatically and at massive scale."
On fully autonomous weapons, Amodei stated: "Today, frontier AI systems are simply not reliable enough to power fully autonomous weapons. We will not knowingly provide a product that puts America's warfighters and civilians at risk." Anthropic clarified it supported "partially autonomous weapons, like those used today in Ukraine," and offered to collaborate with the DoD on R&D to improve reliability for future autonomous applications. The Pentagon rejected that offer.
The Pentagon's position was unambiguous. Spokesman Sean Parnell framed it: "Allow the Pentagon to use Anthropic's model for all lawful purposes. This is a simple, common-sense request." DoD Undersecretary Emil Michael argued: "We can't have a company that has a different policy preference that is baked into the model." A senior Pentagon official told Axios: "It will be an enormous pain in the ass to disentangle, and we are going to make sure they pay a price for forcing our hand like this."
The dispute escalated rapidly through early 2026. On January 12, 2026, Hegseth publicly jabbed Anthropic while announcing Grok's deployment across DoD networks, stating the Pentagon would "not employ AI models that won't allow you to fight wars." Semafor confirmed he was specifically referring to Anthropic — the earliest public reporting of the rift.
On February 15, 2026, Axios broke the exclusive that the Pentagon was threatening to sever its relationship entirely. A senior administration official told Axios "everything's on the table," describing a "broader culture clash" with "the most 'ideological' of the AI labs." The official conceded it would be difficult to quickly replace Claude because "the other model companies are just behind" in specialized government applications.
On February 24, 2026, Hegseth met Amodei at the Pentagon and issued an ultimatum: 5:01 PM on Friday, February 27 to agree to the terms. He threatened to invoke the Defense Production Act and designate Anthropic a supply chain risk. Anthropic's official response noted the contradiction: "These latter two threats are inherently contradictory: one labels us a security risk; the other labels Claude as essential to national security."
On February 26, 2026, Amodei published his definitive statement on Anthropic's blog, titled "Statement from Dario Amodei on our discussions with the Department of War." The key line: "These threats do not change our position: we cannot in good conscience accede to their request." Anthropic also rejected a Pentagon compromise proposal, saying "new language framed as compromise was paired with legalese that would allow those safeguards to be disregarded at will."
When the deadline passed on February 27, Trump posted on Truth Social: "The Leftwing nut jobs at Anthropic have made a DISASTROUS MISTAKE trying to STRONG-ARM the Department of War, and force them to obey their Terms of Service instead of our Constitution. Therefore, I am directing EVERY Federal Agency in the United States Government to IMMEDIATELY CEASE all use of Anthropic's technology." Hegseth formally designated Anthropic a "Supply-Chain Risk to National Security" — the first time this label, previously reserved for foreign adversaries, was applied to an American company.
Anthropic stood alone. Every other major AI lab either accepted the "any lawful use" framework or had already removed prior military restrictions.
OpenAI quietly revised its usage policy on January 10, 2024, removing the blanket prohibition on "military and warfare" use that had been explicit in previous versions. By February 2026, OpenAI secured the classified deployment deal to replace Anthropic on Pentagon networks, announcing the contract within hours of Anthropic's blacklisting. CEO Sam Altman admitted the deal was "definitely rushed" and "the optics don't look good." OpenAI accepted "all lawful purposes, consistent with applicable law" language, though it claimed to maintain technical safeguards. Notably, the Pentagon had already been accessing OpenAI models through Microsoft Azure's enterprise licensing before OpenAI officially lifted its ban — a loophole through Microsoft's terms.
Google underwent the most dramatic reversal. In 2018, after over 4,000 employees signed a petition protesting Project Maven (Pentagon drone surveillance AI), Google withdrew and published AI Principles explicitly banning weapons and surveillance applications. On February 4, 2025, Google deleted the entire "Applications we will not pursue" section from its AI Principles page — removing the prohibitions on weapons, surveillance, and technologies causing "overall harm." Former Google responsible AI director Tracy Pizzo Frey called it: "The last bastion is gone. It's no holds barred." Google's Gemini agreed to the "all lawful use" clause without conditions and now powers the military's GenAI.mil platform.
xAI (Elon Musk's company) accepted the "any lawful use" language without any conditions — the most permissive stance of any major AI company. Grok was approved for classified settings, and its contract was finalized on February 25, 2026, despite the model having previously generated racist and antisemitic content. Meta made its open-source Llama models available to defense agencies and contractors including Lockheed Martin, Booz Allen Hamilton, and Palantir, and partnered with Anduril (a weapons manufacturer) in May 2025. Microsoft, already holding a $9 billion Joint Warfighting Cloud Capability contract, never imposed significant guardrails on military use and had pitched OpenAI's DALL-E as a "battlefield tool."
Gregory Allen of CSIS offered a telling observation: "The user base within the Department of Defense loves Anthropic, loves Claude, and says that their restrictions on usage… have never been triggered."
On March 6, 2026, Anthropic filed a lawsuit alleging First Amendment violations and illegal retaliation. On March 26, 2026, U.S. District Judge Rita Lin in San Francisco issued a 43-page ruling granting a preliminary injunction blocking the supply chain risk designation. Her opinion was blistering: "Nothing in the governing statute supports the Orwellian notion that an American company may be branded a potential adversary and saboteur of the U.S. for expressing disagreement with the government." She found the Pentagon's own records showed it designated Anthropic as a supply chain risk because of its "hostile manner through the press," calling this "classic illegal First Amendment retaliation."
The definitive primary sources for this story, in chronological order:
This was not a routine contract dispute. It was a stress test of whether an AI company's safety commitments survive contact with the most powerful customer on earth. Anthropic's two red lines — no mass surveillance of Americans, no fully autonomous weapons — were modest by most ethical standards, and by the Pentagon's own users' admission, had never actually constrained military operations. Yet every other major AI lab yielded. OpenAI's military ban lasted until it became commercially inconvenient; Google's weapons prohibition survived seven years before being quietly deleted. The federal court's intervention established an important precedent: the government cannot brand a company a national security threat for maintaining terms of service it disagrees with. But the underlying tension — who decides the boundaries of military AI — remains unresolved.