The artificial intelligence company Anthropic was proud to support America’s military. In an essay titled “The Adolescence of Technology,” Dario Amodei, Anthropic’s CEO, wrote that this pride emerged from realizing that “the only way to respond to autocratic threats is to match and outclass them militarily.” He continued, by stating: “The formulation I have come up with is that we should use AI for national defense in all ways except those which would make us more like our autocratic adversaries.”
Anthropic’s AI agent, Claude, has an extensive, 84 page “soul document,” which aims to “avoid large-scale catastrophes,” such as a “world takeover by either AIs pursuing goals of their own that most humans wouldn’t endorse… or by a relatively small group of humans using AI to illegitimately and non-collaboratively seize power.” Claude has a vast amount of applications in the military realm already: either by synthesizing large amounts of intelligence information, or boosting the efficacy of Pentagon hackers — it’s certainly an asset. For instance, Claude was reportedly utilized in the raid to capture now ex-president Nicolás Maduro in Venezuela, through a contract with the data company Palantir Technologies. While Claude’s deployment must comply with its Usage Policies, the intervention was enough to reignite concern in Amodei, who has spoken out before regarding the urgent need for safeguards and guardrails in the use and development of AI.
The Pentagon wasn’t entirely impressed. Chief spokesman Sean Parnell expressed that the Defense Department’s relationship with Anthropic was undergoing review. “Our nation requires that our partners be willing to help our war fighters win in any fight,” he stated. Last summer, Anthropic won a significant endorsement from the Department of Defense when it was awarded a $200 million contract. However, the company’s continued objections to having its technology used in autonomous lethal operations and Amodei’s fears about AI being used as a tool for mass surveillance, have only added more friction to what was arguably an already tense relationship.
The Department of Defense is fully on board with developing autonomous technologies for warfare. After all, for 2026’s fiscal year budget, the agency requested $13.4 billion solely for the continued development of AI-enabled and unmanned systems. Anthropic never claimed that such weapons ought not to exist, but rather, that they should be in the hands of men, not AI. Killer drones might one day be more accurate than a human operator, but Anthropic’s leaders believe that AI has not reached that threshold.
Secretary of Defense Pete Hegseth was seemingly firm on having his way. No compromises, especially not with what he deemed were “woke AI” companies. Anthropic needed to allow the U.S. to use its technologies in all “lawful” purposes, which incidentally included not merely AI-directed warfare, but state-sanctioned surveillance on American citizens.
‘Surveillance’ is a nebulous term — due to a lack of regulation from the Constitution and other statutes, there isn’t a meaningful limit on the amount of digital information that the government may collect. Beyond that, according to Michelle Kim from the MIT Technology Review, the government may purchase commercial data from companies, allowing them to access information that typically wouldn’t be available without a warrant or a subpoena. Someone’s location, their age, email addresses, gender, payment information, and browsing history — all of it is fair game when the government buys the data in bulk. With the helpful assistance of AI, it isn’t exactly difficult to determine someone’s social connections, interests, neuroses, and political views. Along with the location data, it reveals routines and habits. In the words of Amodei, the AI can assemble “a comprehensive picture of any person’s life — automatically and at a massive scale.”
Federal agencies aren’t supposed to collect such intimacies of U.S. citizens in bulk, but have worked around the 2015 change to the Foreign Intelligence Surveillance Act by simply purchasing it directly. According to Kash Patel, the Federal Bureau of Investigation Director, the agency purchases “commercially available information that’s consistent with the Constitution and the laws under the Electronic Communications Privacy Act”, which has “led to some valuable intelligence for us.” These statements were expressed at a Senate hearing mid-March 2026, in response to Senator Ron Wyden’s question on whether Patel would commit to not buying Americans’ location data. The answer seemed to be a resounding no.
One may wonder what exactly about their 2 a.m. browsing habits of pirating Swedish horror movies is so prized. Considering that said data could be used to target journalists, protesters, and immigrants, it would be rather myopic to dismiss the massive data collection as fruitless or inoffensive even if one were innocent. Immigration and Customs Enforcement, after all, signed a contract last year with the company Penlink in order to monitor people’s movements, following them inside their homes, to their workplaces, and beyond.
AI would only supercharge these efforts. Fearing all of this, Anthropic continued to resist. Secretary Hegseth then delivered the ultimatum: Anthropic had until February 27th, 2026, precisely 5:01 p.m. ET, to comply with their demands and sign. The ‘or else’ was far from implied: If they refused to bend, the Trump administration would invoke the Defense Production Act, labeling the company a ‘supply chain risk’ and putting Anthropic’s government contracts in jeopardy.
At 5:14 p.m. ET, Secretary Hegseth accused Anthropic of seeking to “seize veto power over the operational decisions of the United States military.” He officially deemed the company a supply-chain risk, a designation that had never been used before towards an American-based corporation. It’s one that’s typically reserved for companies that have direct connections to foreign adversaries, like China or Russia. This, then, would require all military contractors to prove that their military work holds no ties with Anthropic; and if Secretary Hegseth’s expressions were to be believed, all of those contractors would be compelled to cut commercial ties with the corporation — the latter, of course, is a move not legally substantiated. Anthropic challenged this, and on March 9th, 2026, they sued the federal government in a California court. Anthropic argued that the Trump administration’s actions had caused the corporation irreparable harm, and as such, they requested an injunction of the supply-chain risk label, according to Justin Hendrix of Tech Policy Press. Some weeks later, at a hearing on March 24th, Judge Rita F. Lin stated that she found the federal government’s actions troubling, and that it seemed less about national security concerns and more about the administration punishing Anthropic severely for not bending to their will. Two days later, Judge Lin granted a motion for a preliminary injunction, blocking the Trump administration from enforcing its ban on doing business with the corporation and its designation as a “supply-chain risk.” The Trump administration has made it clear that they will be appealing, but no other statements have been made so far on the matter.
In the meantime, the federal government decided to forge an agreement with a company that they haven’t had any ideological clashes with — one that, on the contrary of Anthropic, openly curries their favor: OpenAI. OpenAI’s CEO Sam Altman shared in a post on X that the agreement would be modified so that their technologies wouldn’t be utilized for surveillance purposes on American citizens. However, many critics remain unswayed. They claim that the revised document still contains heavily ambiguous language and conspicuous exceptions for the utilization of the tech for domestic surveillance by many intelligence agencies within the Department of Defense. Senior counsel for AI Safety, Brian McGrail, for his part, stated that it’s far too common for the intelligence and national security community to interpret carve-outs in a broad fashion. “So could there be some new loophole to be exploited here that we aren’t predicting? It’s totally possible,” he added. After all, there isn’t any clear reassurance on whether the contract is permanently grounded in the current notions of legality. The Trump administration, experts warn, could alter said boundaries via issuing expansive executive orders or legal opinions.
The relationship between Anthropic and the Trump administration continues to be a fraught and thorny subject, as they outwardly appear to profess extremely opposing views on how AI should be handled. Now, a new — or perhaps old — battle appears to be underway, with Anthropic calling for putting the brakes on reaching the final frontier of AI development, and the Trump administration being eager to push towards deregulation. “Whoever wins AI, wins,” said the President, in response to the domestic backlash. We have to do everything in our power to fight against those autocratic threats, after all.