Claude has just been ejected from the entire American administration. After a months-long standoff with the Pentagon, Anthropic refused to cross two ethical boundaries that Washington demanded be removed, and Donald Trump responded with a total decree.
From Strategic Partner to State Enemy
Everything had started well, however. In 2024, Anthropic signed a 200 million dollar contract with the Department of War, displaying a certain pride at the time: it was the first cutting-edge AI company to operate on classified American networks. Claude, its flagship model, was deployed via Palantir, the firm known for having helped locate Osama bin Laden. This exclusive position was supposed to be a springboard. It has become the central point of friction in an unprecedented political crisis.
The break crystallized in January 2026 around a specific episode. On January 3, during the capture of Venezuelan President Nicolás Maduro, the Pentagon used Claude through Palantir's programs, revealed the Wall Street Journal on February 15. Anthropic raised questions about this use. The Pentagon, furious at these ethical hesitations, began to prepare its response.
Two red lines the Pentagon wanted to erase
The conflict concerns two specific restrictions that Anthropic has imposed in its contracts since the beginning: no use of Claude for fully autonomous weapons, and no use for mass domestic surveillance of American citizens. The Pentagon demanded unrestricted access, invoking the right to use the models for "all lawful purposes." Total impasse.
Pete Hegseth, Secretary of War, summoned CEO Dario Amodei on Tuesday, February 25, and gave him an ultimatum: lift these restrictions before Friday, February 27, at 5:01 PM, under penalty of sanctions. Amodei's response came on Thursday afternoon, in a long statement. "These threats do not change our position: we cannot in good conscience accede to their demand," he wrote in this official statement. (These threats do not change our position: we cannot in good conscience accede to their demand.) He also explained that fully autonomous weapons "must be deployed with appropriate safeguards, which do not currently exist."
Amodei's position is not anti-militarism. He explicitly clarified this: "I deeply believe in the existential importance of using AI to defend the United States and other democracies, and to defeat our autocratic adversaries," he wrote in the same statement, before adding that "in certain specific cases, we believe AI can undermine, rather than defend, democratic values."
A designation reserved for foreign adversaries
The Trump administration's reaction went far beyond a simple contract cancellation. Pete Hegseth designated Anthropic as supply chain risk of the American defense and prohibited any military contractor from working with it. Donald Trump then ordered on Truth Social, in all caps, that all federal agencies "IMMEDIATELY CEASE ANY USE OF ANTHROPIC TECHNOLOGY." A six-month transition period is granted to departments already integrated.
This type of designation had previously been reserved for companies like Huawei or ZTE, Chinese companies considered under the influence of a foreign government. Anthropic is a startup founded in San Francisco, partially funded by Amazon. Several American legal experts have already called this decision a dangerous precedent: nothing in the texts prohibits the administration from using this mechanism against any national tech company that refuses to comply with its contractual requirements. The company announced it will challenge the designation in court, relying on Article 3252 of Title 10 of the US Code, which also limits its scope to contracts of the Department of War only.
Cloud giants in uncertainty
Hegseth's wording poses a much broader problem than just the Anthropic case. Any entity "doing business with the military" would no longer be able to work with the startup. However, Claude runs on the infrastructure of Amazon Web Services, Microsoft Azure, and Google Cloud, which also hosts some of the largest American defense contracts.
If we stick to a strict interpretation, these partnerships would be threatened. No one has yet pushed the logic that far, but the wording is indeed in the official text. For an administration that has just shown it can use regulatory designations as leverage, the wording is not insignificant. Pentagon officials had indeed contacted groups like Lockheed Martin and Boeing as early as February to assess the extent to which they use Claude, according to Wall Street Journal.
OpenAI and xAI pick up the pieces
February 25th, the day of the ultimatum to Anthropic, was also the day of a signing. xAI, Elon Musk's company, concluded an agreement with the Pentagon to deploy its AI Grok on classified military networks, thus breaking Claude's exclusive position on these networks. xAI has unconditionally accepted the standard of use for "all lawful purposes," the exact wording Anthropic resisted.
OpenAI is also on the verge of reclaiming the position left vacant by Anthropic within the Pentagon, according to France 24. The sequence is clear: while Anthropic refused to yield, its direct competitors filled the void left, without imposing ethical conditions. Dario Amodei had nevertheless said he hoped the Pentagon would reverse its decision, while assuring that Anthropic would "strive to ensure a smooth transition to another provider" if the contract were canceled.
Dario Amodei faces Washington's contradictions
The CEO of Anthropic also pointed out a major inconsistency in the administration's position. On one hand, Hegseth designates Anthropic as national security risk. On the other, Trump threatens with "severe civil and criminal consequences" if the transition is not properly managed, implying that Claude remains temporarily indispensable. "These two threats are inconsistent: one labels us a security risk, the second makes Claude an essential element for national security," Amodei wrote in his official statement.
In this tense context, unexpected solidarity emerged: employees from Google and OpenAI publicly supported Anthropic against the administration. This signal illustrates that the AI sector as a whole is closely monitoring a precedent that could, tomorrow, concern any other company refusing to align its practices with governmental requirements.
Eighteen months after presenting its contract with the Pentagon as a victory, Anthropic finds itself in legal dispute with Washington, deprived of its main government outlets, but with its principles intact. The case goes far beyond the start-up itself: it raises a structural question about the conditions under which American AI companies can, or cannot, impose ethical limits on their sovereign clients. The legal battle opening around section 3252 of title 10 of the US Code will be closely watched by the entire sector.
No comments yet — start the discussion!