President Trump is terminating the government’s relationship with Anthropic, an AI company whose products, until recently, were used by Pentagon officials for classified operations. Following a weekslong standoff with the company, Trump posted on Truth Social this afternoon that all federal agencies must “IMMEDIATELY CEASE all use of Anthropic’s technology,” adding: “We don’t need it, we don’t want it, and will not do business with them again!” The General Services Administration announced that it would take action against Anthropic’s products, and indeed, according to an email I obtained that was sent to the leadership of all agencies using USAi—a GSA platform that provides chatbots from tech companies to government workers—access to Anthropic was suspended “immediately.” The government is also removing Anthropic from its primary procurement system, which is the key way for any federal agency to purchase a commercial product.
Anthropic was awarded a $200 million contract with the Pentagon last summer geared toward providing versions of its technology for military use. OpenAI, Google, and xAI were awarded similar contracts, though Anthropic’s Claude models are the only advanced generative-AI programs to receive Pentagon security clearance permitting the handling of secret and classified data. Claude had been integrated across the Department of Defense and was reportedly used to assist the raid on Venezuela that led to the capture of President Nicolás Maduro.
Anthropic has said that it will not allow Claude to be used for mass domestic surveillance or to enable fully autonomous weaponry, which could involve applications such as Claude selecting and killing targets with drones, and analyzing data that have been indiscriminately gathered on Americans by the intelligence community. Anthropic has also said that the Pentagon never included such uses in its contracts with the firm. But now DOD is demanding unrestricted use of Claude and accusing Anthropic of trying to control the military and “putting our nation’s safety at risk” by refusing to comply.
Following a heated meeting on Tuesday, DOD gave Anthropic until today at 5:01 p.m. eastern time to acquiesce to its demands. If not, the Pentagon would compel the company under an emergency wartime law called the Defense Production Act or, even more severe, designate Anthropic a “supply-chain risk,” which could forbid any organization that works with the U.S. military to do business with the AI company. Shortly after Trump’s announcement, Defense Secretary Pete Hegseth declared that he was doing just that. Dean Ball, an analyst who helped write some of the Trump administration’s AI policy, has called the threats “the most aggressive AI regulatory move I have ever seen, by any government anywhere in the world.”
Last night, Anthropic CEO Dario Amodei wrote in a public letter, “We cannot in good conscience accede to” the Pentagon’s request. Following Trump’s and Hegseth’s orders today, Anthropic said in a statement, “No amount of intimidation or punishment from the Department of War will change our position.” DOD, which the Trump administration refers to as the Department of War, did not immediately respond to requests for comment.
The situation signals a potentially seismic shift in relations between Silicon Valley and the federal government. Defense officials and technology companies alike are concerned that the U.S. military is losing its technological edge over its adversaries, particularly China—in part because the private sector, rather than the Pentagon, is where much American innovation comes from these days. And instead of federal grants, the massive investments needed for generative AI have come from tech companies themselves. Historically, companies the Pentagon works with have not set terms for how the government uses their products. But as Thomas Wright recently wrote in The Atlantic, this dynamic is complicated when it comes to AI tools made fully by a private sector that understands the technology far better than the government does.
Anthropic has shown itself to be eager to work with the government and the military, hence it being the first of the frontier AI firms to receive such a high security clearance from the military. Amodei is by far the most hawkish of any prominent AI executive, warning frequently about the need for democracies to use AI to vanquish authoritarianism and, especially, stay ahead of China. In the letter he published last night, Amodei wrote: “I believe deeply in the existential importance of using AI to defend the United States and other democracies, and to defeat our autocratic adversaries.” And although he took a principled stance against domestic surveillance, Amodei wrote that he is open to Claude eventually being used to power fully autonomous weapons—just not yet, because today’s best AI models “are simply not reliable enough” to do so. Developing such AI-powered weapons in the present, he wrote, would put American soldiers and civilians at risk.
Much remains uncertain about the unraveling relationship between the Trump administration and Anthropic, but the White House has been souring on Anthropic for months. Amodei has been publicly critical of Trump, and wrote a lengthy Facebook post in support of Kamala Harris during the 2024 election. White House officials have called the company “woke” and accused it of “fear mongering.”
We have ended up in a paradoxical situation in which the U.S. government is at once saying that Claude is so essential to national security that it could invoke an emergency law to exert extensive control over Anthropic and that the company is so woke and radical that using Claude would itself be a national-security risk. “I don’t understand it,” a former senior defense official who requested anonymity to speak freely told me. “It’s an existential risk if you use it or if you don’t.”
Many in Silicon Valley have rallied in support of Anthropic, even as the major companies have maintained their business with the government. (The precise terms of the Pentagon’s contracts with other AI companies have not been made public.) Jeff Dean, a top Google executive, wrote on X that generative AI should not be used for domestic mass surveillance. OpenAI CEO Sam Altman wrote in an internal memo circulated last night, a copy of which I obtained, that “we have long believed that AI should not be used for mass surveillance or autonomous lethal weapons,” and he has expressed similar sentiments publicly. More than 500 current employees of both OpenAI and Google—many of them anonymous—signed an open letter in support of Anthropic. On the sidewalk outside Anthropic’s headquarters in San Francisco today, passersby scribbled messages of support with chalk.
The fallout from the supply-chain-risk designation is still unclear. In theory, Google, Microsoft, Amazon, and several other behemoths that contract with the federal government will have to stop doing business with Anthropic, which would be a mess for everyone involved and potentially devastating for Anthropic; Amazon, for instance, is building data centers that will train future versions of Claude. But just how sweeping of an impact such a designation would have on Anthropic’s customers is up for debate, and the company said in its statement today that many applications of Claude, even for customers that partner with DOD, will not be affected.
Meanwhile, private AI firms will continue to be important to the federal government as it works to compete with China, Russia, and all manner of adversaries. Trump gave the Pentagon six months to phase out Claude, which suggests that the technology has indeed become essential—and is essential to replace. And at some point, the U.S. military may no longer find itself in a position to dictate its terms. Altman, in his internal memo, wrote that OpenAI is exploring a contract with the Pentagon to use its AI models for classified workloads that would still exclude uses that “are unlawful or unsuited to cloud deployments, such as domestic surveillance and autonomous offensive weapons.” The Pentagon reportedly agreed to those conditions shortly after announcing that it would sever ties with Anthropic, although no contract has been signed. But other figures in tech, including the Anduril co-founder Palmer Luckey and the investor Katherine Boyle, have come out in support of demands for unrestricted use. This showdown was between the Pentagon and Anthropic. The next may be a war within Silicon Valley itself.
Great Job Matteo Wong & the Team @ The Atlantic for sharing this story.
