The AI Company That Refused the Pentagon — And What Happened Next
On the evening of February 27, 2026 — eight hours before the United States launched airstrikes on Tehran — President Trump ordered every federal agency to immediately stop using Anthropic’s AI products. Defense Secretary Pete Hegseth designated Anthropic a “supply chain risk to national security,” a classification previously reserved for foreign adversaries like Huawei.
The reason: Anthropic refused to let the Pentagon use its AI model Claude for fully autonomous weapons or mass domestic surveillance of American citizens.
The conflict between a trillion-dollar defence apparatus and an AI safety company over two usage restrictions has become the most consequential AI policy story of 2026. Here’s the full story — and why it matters far beyond just one company.
How It Started: The $200 Million Contract
In 2024, Anthropic embedded Claude into classified networks through Palantir, and the arrangement worked. By July 2025, the Pentagon signed the $200 million contract, fully aware of Anthropic’s usage restrictions. Claude became the first frontier AI model deployed in the US government’s classified networks — used for intelligence analysis, operational planning, cyber operations, and modelling and simulation.
The arrangement was mutually beneficial and, by all accounts, operationally successful. Claude is extensively deployed across the Department of War for mission-critical applications, such as intelligence analysis, modelling and simulation, operational planning, cyber operations, and more.
The fracture began in January 2026, when US special operations forces captured President Nicolás Maduro in Caracas. Eighty-three people were killed, including 47 Venezuelan soldiers. In mid-February, media reports revealed that Claude had been used during the active operation through Anthropic’s partnership with Palantir. An Anthropic executive contacted Palantir to ask whether the technology had been used in the raid.
The Pentagon interpreted this inquiry as Anthropic seeking approval authority over military operations. Anthropic denied this interpretation. The relationship deteriorated rapidly from that point.
Timeline
Hegseth issues AI strategy memo
Defense Secretary directs all DoD AI contracts adopt “any lawful use” language — removing Anthropic’s two restrictions. Negotiations begin between Pentagon and Anthropic.
Formal ultimatum delivered
Defense Secretary Hegseth delivers a formal demand to Anthropic CEO Dario Amodei: remove all usage restrictions or face termination of the $200 million contract, designation as a supply chain risk, or potential invocation of the Defense Production Act.
Anthropic refuses. Deadline expires.
Anthropic holds its two restrictions. Hegseth announces on X that Anthropic is designated a supply chain risk. Trump orders all federal agencies to cease using Anthropic’s technology.
Airstrikes on Iran — using Claude
The Wall Street Journal reported that US strikes in Iran used Anthropic’s technology hours after Trump announced the ban. The government had designated Claude a national security risk while simultaneously relying on it in active combat operations.
OpenAI announces a Pentagon deal
OpenAI announced it had reached a deal that will allow the US military to use its technologies in classified settings. CEO Sam Altman acknowledged the negotiations were “definitely rushed.” OpenAI’s deal includes softer protections against autonomous weapons and surveillance than Anthropic’s hard limits.
Defense companies begin switching off Claude
Defense tech companies are telling employees to stop using Claude and to switch to other AI models. Ten portfolio companies of one major defense tech VC have backed off their use of Claude for defense use cases. Palantir — whose operations are heavily intertwined with Claude — faces near-term disruption.
Anthropic’s Position — In Its Own Words
Anthropic’s public response was measured but firm. The company said it had tried in good faith to reach an agreement with the Pentagon, making clear that it supports all lawful uses of AI for national security aside from the two narrow exceptions. “To the best of our knowledge, these exceptions have not affected a single government mission to date.”
On the autonomous weapons restriction specifically, Anthropic’s reasoning was technical, not political: “We do not believe that today’s frontier AI models are reliable enough to be used in fully autonomous weapons. Allowing current models to be used in this way would endanger America’s warfighters and civilians.”
Anthropic also noted its national security track record: it was the first frontier AI firm to deploy on classified networks, cut off CCP-linked firms at a cost of hundreds of millions in revenue, and shut down a CCP-sponsored cyberattack that attempted to abuse Claude.
The Legal Problem With What the Pentagon Did
Legal analysts have been almost uniformly sceptical of the Pentagon’s action. The designation exceeds what the statute authorizes. The required findings don’t hold up. Hegseth’s own public statements may have doomed the government’s litigation posture before it even begins.
The most glaring contradiction: Hegseth has declared it safe to leave Anthropic integrated into military networks for another six months for “a seamless transition.” The government cannot simultaneously claim a vendor poses an acute supply chain threat requiring emergency exclusion and that it’s perfectly safe to keep using the vendor for half a year — or, apparently, for active combat operations.
What This Means for Everyone in Tech
The Anthropic-Pentagon story is not just a story about one AI company and one government contract. It’s a preview of a question the entire industry will have to answer: when does an AI company’s responsibility end and a government’s authority begin?
The Pentagon’s move is a signal to other AI companies looking to make millions selling their services to the government — to make sure they do not attempt to put any sort of restrictions on AI’s uses.
OpenAI’s rushed deal — which gives the Pentagon access with softer legal protections rather than Anthropic’s hard ethical ones — is now the template being cited. Whether those softer protections hold under operational pressure remains to be seen.
The final irony is perhaps the most pointed: the AI model that the US government deemed a national security risk was, on the same night, reportedly helping plan its strikes on Iran. Whatever you think of Anthropic’s position, that contradiction alone suggests the designation was more political theatre than genuine security policy.