Major Silicon Valley AI firm Anthropic has walked away from talks with the U.S. Department of Defense and released a statement from CEO Dario Amodei publicly opposing the federal government, in response to Anthropic’s concerns over how they will use Anthropic’s products both in foreign wars and for domestic purposes. The federal government has responded aggressively, with President Donald Trump giving the U.S. military six months to stop using Anthropic products entirely.
The standoff between Anthropic and the Pentagon began in earnest on February 24, when Secretary of Defense Pete Hegseth, in a meeting with Amodei at the Pentagon, issued an ultimatum towards Anthropic to either allow the U.S. military unfettered access to Claude AI, Anthropic’s flagship AI model, by February 27, or face severe penalties, likely in the form of termination of a $200 million contract granted to Anthropic and three other AI companies—OpenAI, Google, and xAI— in July of last year. Anthropic’s statement came two days after this meeting, announcing their refusal to comply with the administration.
In the statement, Amodei lists two main concerns: “mass domestic surveillance” and “fully autonomous weapons”. Although not explicitly mentioned in the statement, the former concern could be referring to the use of AI in aiding immigration enforcement. ICE is gradually increasing its utilization of AI in several aspects of its operations, most notably working with data analytics firm Palantir for data extraction, tip processing, and software development.
The second concern, the U.S.’s use of “fully autonomous weapons,” likely relates to the then-increase in deployment of U.S. forces around Iran, which has since developed into a full-scale war that has resulted in the death of Iranian Supreme Leader Ali Khamenei. Claude AI has been used by the U.S. military in the joint U.S.-Israeli bombardment of Iran, even after the Anthropic ban.
Anthropic’s concerns also come in the wake of Claude AI being used by the government in the operation to capture Venezuelan President Nicolás Maduro in January. U.S. special operations forces used Claude both during the planning and the execution of the raid, and it is unclear whether Anthropic’s permission was obtained before the raid occurred or whether the raid violates Anthropic’s usage policy.
On February 27, the day of the deadline passing, the Pentagon designated Anthropic a supply chain risk, meaning that Anthropic would be completely excluded from defense contracts with the U.S. government. In response, Amodei announced in another statement that Anthropic would challenge the ruling in court.
The Trump administration and the Pentagon have made several harsh statements against Anthropic and Amodei since Amodei posted his first statement. In a post on Truth Social, President Trump himself referred to Anthropic as a “radical left, woke company” and accused them of trying to “strong-arm” the Department of DefenseWar, while Hegseth commented on X, describing Anthropic’s behaviour as a “master class in arrogance and betrayal,” additionally stating that “the Department of War must have full, unrestricted access to Anthropic’s models for every LAWFUL purpose in defense of the Republic.”Meanwhile, Anthropic rival OpenAI has made its own agreement with the federal government on February 27, allowing them access to classified systems, although stating that its terms of service also prohibited mass domestic surveillance and fully autonomous weapons. However, the company amended the agreement a week later due to severe backlash from their consumer base, with CEO Sam Altman making an apology and saying the move “looked opportunistic and sloppy.”
