Anthropic, the maker of the Claude series of large language models, had a contract with the Department of Defense to be the first LLM provider authorized to work on classified systems. When the Wall Street Journal reported that Claude was used to sort through data collected by Palantir to find targets to bomb in the raid on Venezuela, Anthropic contacted Palantir on how exactly Claude was used.

Anthropic is a strong proponent of artificial intelligence safety and responsible use, performing large amounts of research and studies on model behavior and defenses against jailbreaks and safeguard systems. This sits in stark contrast to many other LLM developers, such as OpenAI’s sycophantic GPT-4o model that users developed romantic attraction to, or xAI’s Grok model which has called itself MechaHitler.

Part of this strong stance on safety is their strict Acceptable Use Policy, which includes that Anthropic’s LLM models cannot be used for mass surveillance, that Claude cannot be used for violent actions and that a human must make the final decision on high-risk use cases. Anthropic worried that the U.S. Department of Defense violated the latter two policies.

When the $200 million contract was first awarded, relations between the Trump administration and Anthropic were already sour, as the “no mass surveillance” policy went at odds with the administration’s stance on how technology can be used by federal departments. Secretary of Defense Pete Hegseth made jabs at Anthropic’s restrictions during a press conference announcing that xAI’s Grok series of LLMs would be authorized for use in the military.

This relationship further soured when Anthropic questioned Palantir on their use of Claude in the Venezuelan raid, and the DoD immediately took the nuclear option: they threatened to label Anthropic a supply chain risk and ban their use in government contracts if they did not remove any and all safeguards and usage restrictions on Claude.

When the deadline of Feb. 27 passed, the DoD followed through on their threat, labeling Anthropic a supply chain risk.

This week, Anthropic delivered a master class in arrogance and betrayal as well as a textbook case of how not to do business with the United States Government or the Pentagon.

Our position has never wavered and will never waver: the Department of War must have full, unrestricted…

— Secretary of War Pete Hegseth (@SecWar) February 27, 2026

Although Hegseth’s statement on X, formerly known as Twitter, states that “effective immediately, no contractor, supplier, or partner that does business with the United States military may conduct any commercial activity with Anthropic,” under 10 USC 3252, this restriction only applies to use in contracts with the military, not in general.

In a statement by Anthropic before the deadline, this fact was highlighted, with Anthropic stating, “If you are an individual customer or hold a commercial contract with Anthropic, your access to Claude—through our API, claude.ai, or any of our products—is completely unaffected. If you are a Department of War contractor, this designation—if formally adopted—would only affect your use of Claude on Department of War contract work. Your use for any other purpose is unaffected.”

OpenAI, creators of the competing GPT series of models, almost immediately got a contract with the DoD to replace Anthropic in classified uses. Although CEO Sam Altman would later state that they were able to change the contract to include the same or higher restrictions as Anthropic’s contract, the initial contract signed did not include any such restrictions.

This sparked outrage amongst users, who felt that OpenAI was “war profiteering,” leading to an estimated four million users canceling their subscriptions in protest, according to activist group QuitGPT. Anthropic has massively benefited from this boycott, with users migrating to the Claude series of models, and Anthropic adding new features to make it easier to do so.

Anthropic has filed a lawsuit against the DoD to block the supply chain risk label. 

“We do not believe this action is legally sound, and we see no choice but to challenge it in court,” CEO Dario Amodei said.

Researchers and engineers at OpenAI and Google DeepMind have filed an amicus brief in support of Anthropic. In a press release by OpenAI about their DoD contract and their updated usage restrictions, they stated that they don’t know why Anthropic could not reach an agreement for usage restrictions.

Despite Anthropic being labeled a supply chain risk, the United States armed forces continue to use Claude in their operations through the 60-day grace period, particularly in Iran. When an Iranian girls’ school was destroyed by Tomahawk cruise missiles, Reuters reported that outdated and unverified intelligence was to blame for the mistaken targeting. This Week in Worcester spoke to various insider sources in the DoD who state that the errors in intelligence and targeting were caused by or exacerbated by the use of AI systems such as Claude. 

“The immediate theory is that the AI program included the school’s position based on older, archived intelligence. The logic behind the launch, and the mechanics of who authorized it is unclear,” an anonymous Department of Justice appointee said.

According to the sources, the DoD is rapidly scaling up its use of a Claude-based system over the past year, integrating it with many core operational decisions.

“They are gung-ho about this program and want to use it for everything. Most of their operational planning is done using this software, although there is some things we have designed in-house,” an anonymous Department of Defense logistics programmer said.