US labels Anthropic a supply chain risk in clash over military AI use
The Trump administration’s unprecedented move against Anthropic over AI safeguards is forcing government contractors to reconsider their use of the company’s chatbot Claude.
The US administration is following through with its threat to designate artificial intelligence (AI) company Anthropic as a supply chain risk in an unprecedented move that could force other government contractors to stop using the AI chatbot, Claude.
The Pentagon said in a statement Thursday that it has “officially informed Anthropic leadership that the company and its products are deemed a supply chain risk, effective immediately.”
The decision appeared to shut down the opportunity for further negotiation with Anthropic, nearly a week after President Donald Trump and defence secretary Pete Hegseth accused the company of endangering national security.
Trump and Hegseth announced a series of threatened punishments last Friday, on the eve of the Iran war, after Anthropic CEO Dario Amodei refused to back down over concerns the company’s products could be used for mass surveillance of Americans or autonomous weapons.
Amodei said in a statement Thursday that “we do not believe this action is legally sound, and we see no choice but to challenge it in court.”
"This has been about one fundamental principle: the military being able to use technology for all lawful purposes," the Pentagon statement said.
"The military will not allow a vendor to insert itself into the chain of command by restricting the lawful use of a critical capability and put our warfighters at risk,“ it added.
Amodei countered that the narrow exceptions Anthropic sought to limit surveillance and autonomous weapons “relate to high-level usage areas, and not operational decision-making”.
He added there were “productive conversations” with the Pentagon in recent days over whether it could keep using Claude or establish a “smooth transition” if no agreement was reached.
Trump gave the military six months to phase out Claude, which is already widely embedded in military and national security platforms. Amodei said it’s a priority to make sure warfighters won’t be “deprived of important tools in the middle of major combat operations.”
Some military contractors were already cutting ties with Anthropic, a rising star in the tech industry that sells Claude to a variety of businesses and government agencies.
Defence company Lockheed Martin said it will “follow the President’s and the Department of War’s direction” and look to other providers of large language models.
“We expect minimal impacts as Lockheed Martin is not dependent on any single LLM vendor for any portion of our work,” the company said.
How the US Defence Department will interpret the scope of the risk designation is unclear. Amodei said a notification Anthropic received from the Pentagon on Wednesday shows it only applies to Claude's use by customers as a “direct part of” their military contracts.
Microsoft said its lawyers studied the rule and the company “can continue to work with Anthropic on non-defence related projects.”
Pentagon draws criticism for its decision
The Pentagon's decision to apply a rule designed to address supply threats posed by foreign adversaries was met with broad criticism. Federal codes have defined supply chain risk as a “risk that an adversary may sabotage, maliciously introduce unwanted function, or otherwise subvert” a system in order to disrupt, degrade or spy on it.
US Senator Kirsten Gillibrand, a New York Democrat and member of the Senate Armed Services Committee and Senate Intelligence Committee, called it “a dangerous misuse of a tool meant to address adversary-controlled technology.”
“This reckless action is shortsighted, self-destructive, and a gift to our adversaries,” she said in a written statement Thursday.
Neil Chilson, a Republican former chief technologist for the Federal Trade Commission who now leads AI policy at the Abundance Institute, said the decision looks like “massive overreach that would hurt both the US AI sector and the military’s ability to acquire the best technology for the US warfighter.”
Earlier in the day, a group of former defence and national security officials sent a letter to US lawmakers expressing “serious concern” about the designation.
“The use of this authority against a domestic American company is a profound departure from its intended purpose and sets a dangerous precedent,” said the letter from former officials and policy experts, including former CIA director Michael Hayden and retired Air Force, Army and Navy leaders.
They added that such a designation is meant to “protect the United States from infiltration by foreign adversaries — from companies beholden to Beijing or Moscow, not from American innovators operating transparently under the rule of law. Applying this tool to penalise a US firm for declining to remove safeguards against mass domestic surveillance and fully autonomous weapons is a category error with consequences that extend far beyond this dispute.”
Anthropic sees boost in consumer downloads
While losing big partnerships with defence contractors, Anthropic experienced a surge of consumer downloads over the past week due to people siding with its moral stance. More than a million people signed up for Claude each day this week, the company said, lifting it past OpenAI's ChatGPT and Google's Gemini as the top AI app in more than 20 countries in Apple's app st
The dispute with the Pentagon has also further deepened Anthropic's bitter rivalry with OpenAI, which started when ex-OpenAI leaders, including Amodei, started Anthropic in 2021.
Hours after the Pentagon punished Anthropic last Friday, OpenAI announced a deal to effectively replace Anthropic with ChatGPT in classified military environments.
OpenAI said it sought similar protections against domestic surveillance and fully autonomous weapons, but later had to amend its agreements, leading CEO Sam Altman to say he shouldn't have rushed a deal that “looked opportunistic and sloppy.”
Amodei also expressed regret about his own part in that “difficult day for the company,” saying Thursday he wanted to “directly apologise” for an internal note he sent to Anthropic staff that attacked OpenAI's behaviour and suggested Anthropic was being punished for not giving ”dictator-like praise" to Trump.
AI on the battlefield: How is the US integrating AI into its military?
The US military used Anthropic's Claude AI, but after Anthropic refused to remove guardrails against mass surveillance and autonomous weapons, the Pentagon cancelled the contract and turned to OpenAI.
Media reports that the US military used Anthropic’s artificial intelligence (AI) chatbot Claude during operations targeting leaders in Venezuela and Iran are raising new questions about how quickly artificial intelligence (AI) is being integrated into warfare
American media reported that Claude was used to help facilitate a January operation that led to the capture of Venezuela’s leader Nicolás Maduro. Similar reports later emerged that the chatbot was also used during preparations for an operation targeting Iran’s deceased supreme leader, Ayatollah Ali Khamenei.
Experts say the incidents offer a rare glimpse into how advanced AI systems may already be supporting US military planning and intelligence work.
“It was very surprising to see the sudden deployment of these tools, especially when I think the larger community does not think that they’re ready for said deployment,” said Heidy Khlaaf, chief AI scientist at US policy thinktank, AI Now Institute.
“We’re sort of questioning whether these AI models can be successful in any military settings at all because of how flawed they are,” she told Euronews Next.
Khlaaf said researchers have warned that large language models can produce unreliable or incorrect outputs, raising concerns about how they might perform in high-stakes environments such as military operations.
The reported use of Claude also comes as the Trump administration pushes an ambitious strategy to make the US military “AI-first”, arguing that rapid adoption of the technology is necessary to compete with rivals such as China.
‘We see this sense of urgency’
The United States has used various forms of automation technology in the military since the 2010s, and it has been a focus area for several presidents, including Trump’s predecessors, Barack Obama and Joe Biden, experts told Euronews Next.
Early AI models were used for logistics, maintenance, or translations, according to Elke Schwarz, a professor of political theory at Queen Mary University of London in the United Kingdom.
Trump’s second mandate accelerates the adoption of generative artificial intelligence (AI) models like OpenAI’s ChatGPT or Anthropic’s Claude in an “AI-arms race,” against the country’s adversaries, both Schwarz and Khlaaf said.
America’s policies give a “sense of urgency” to develop AI because it is a “very valuable technology” that will keep the country ahead of its rivals, said Giorgos Verdi, policy fellow at the European Council of Foreign Relations think tank.
The Department of War’s AI Acceleration strategy aims to secure American military dominance by eliminating barriers to AI integration and investing in strategic projects that will keep the military ahead of rivals.
“The idea really is to bring AI into all kinds of domains, including the harmless ones, but also the more harmful ones,” Schwartz said.
He noted that previous administrations were more cautious about establishing safety guardrails governing how and when such technologies could be used.
As part of that effort, the acceleration strategy has a database called genai.mil, which allows bureaucrats to access AI chatbots, including Google’s Gemini and xAI’s Grok.
The administration’s 2025 budget, called the “Big Beautiful Bill,” also includes hundreds of millions of dollars in funding for AI-related military projects.
The document sets aside $650 million (€550million) for military innovation, including $145 (€123 million) to develop AI-powered counter-drone systems.
Another $250 million would go toward the “advancement of the AI ecosystem,” while a further $250 million is allocated to expand artificial intelligence capabilities at a Cyber Command centre. An additional $115 million is earmarked for accelerating nuclear national security missions using AI.
US military still in ‘trial phase’ with AI chatbots
Due to the“inherent lack of transparency,” it is difficult to determine how advanced the US government is in its plans, Schwarz said. “Unlike a certain ammunition or a specific physical weapon system, you don’t really see what is being used,” she said. “Everything happens in an interface and very much in the zone of invisibility.”
Schwarz believes that the US military is in a “trial phase,” where it is experimenting with different AI companies to understand what they can do and where their limitations lie.
Anthropic’s $200 million partnership with the US military is for a two-year prototype that will advance national security, the company says. It will work with the department to “anticipate and mitigate potential adversarial uses of AI,” and identify any risks with adopting the technology throughout the “defence enterprise.”
Schwartz said this suggests that the systems are being tested in live environments, which she said raises ethical concerns.
“This isa terrible practice for something that involves human lives,” she said.
However, Verdi believes that systems like Claude that were used in the Venezuelan and Iranian contexts for “more mundane tasks,” such as collecting or analysing satellite images.
“A human may not be able to analyse every single piece of intelligence coming in. That’s what the AI system will be able to do more quickly,” he said. “Then, the humans interpret the outputs of the AI system and then act.”
Experts warn of growing interest in AI-powered autonomous weapons
The researchers worry that the growing role of AI in US military planning and decision-making could eventually lead to the development of autonomous weapons.
“I think there is definitely an interest to at least have the option to develop fully autonomous AI-enabled weapons and potentially make use of those,” Verdi said.
Autonomous weapons could be any weapon that could identify, select, and engage with a target without having a human involved in the final decision, Khlaaf said.
“So instead of taking a recommendation from a large language model and a human acting on it or choosing not to, you would then have that be completely automated away,” she said.
One of the main arguments for developing such systems is the fear that the US could fall behind if a rival builds them first, Verdi said.
However, there is no public information suggesting that China has integrated AI in any way into its military, Verdi and Khlaaf said.
The Chinese are “very concerned about keeping that technology under control,” Verdi added.
The AI capabilities of other American opponents, such as Russia, Iran or North Korea, are “even less sophisticated,” Verdi added, so it is even less likely that those countries would have AI autonomous weapons.
Creating fully autonomous weapons with AI can also lead to escalation in a conflict, Verdi said.
A recent pre-print study from King’s College London found that AI chatbots almost always chose to threaten nuclear weapons use in a war game scenario.
Pentagon faces ‘challenging transition’ away from Claude
Verdi said that we should expect the US to continue using Claude or another AI chatbot in their operations because both Venezuela and Iran were “seemingly very effective,” in fulfilling the mission’s objectives.
The perceived success of these missions creates a risk that the US will want to drop even more guardrails, such as human oversight, to make the technology even more effective, she added. The challenge for the Department of War will be to find a model that works as well as Claude, Verdi said.
The government will be phasing it out in the next six months since the company refused to give the military unfettered access to its technology for what Anthropic claims could be used for mass surveillance or autonomous weapon development, according to a statement from Pete Hegseth, the US Secretary of War.
Anthropic said Claude has been rolled out throughout the US government's classified information networks, deployed at national nuclear laboratories, and does intelligence analysis directly for the Department of War.
Meanwhile, the Department of War signed a contract with OpenAI to integrate “advanced AI systems in classified environments,” hours after the Anthropic deal was scrapped.
“I think the Department of War will be looking at a challenging transition, but at the same time, it is not an impossible task,” to replace Claude with a new AI system, he said.
The intelligence collected and provided by Claude will likely stay with the department and could be used by the next provider, he said.
