Even though President Donald Trump last week directed government agencies to “immediately cease” all use of Anthropic’s Claude artificial intelligence tools, they were still being used to launch attacks against Iran.
Claude is deeply integrated into the U.S. military and is the only AI technology approved for classified use, used to plan operations, collect and collate information, identify targets, simulate combat scenarios, and conduct cyber operations.
Claude, which was also used to kidnap Venezuelan President Nicolas Maduro, is deeply entrenched within the U.S. Department of Defense (Pete Hegseth’s “Department of the Army”), and despite President Trump’s immediate directives, the administration gave the Pentagon six months to phase it out and replace it with a human rival model.
The conflict between the Department of Defense and Anthropic is related to the department’s efforts to rewrite contracts it has with AI companies.
The company’s co-founder Dario Amodei has been outspoken about the need for AI regulation, insisting on two red lines that should not be crossed. The company’s model would not be used to conduct mass surveillance of Americans or control autonomous weapons without humans being involved in the “kill chain.”
Both of these applications appear to be illegal, and given that the Department of Defense claims it does not use AI illegally, what’s the problem?
The Department of Defense presents its position as a matter of principle. The company does not want private contractors to limit the use of the products it buys, and claims that the companies that supply the missiles are not allowed to tell the company where it can and cannot deploy its missiles.
Anthropic believes that AI is too immature and dangerous to be deployed without the limits it claims, and argues that current laws have not kept up with the rate at which AI capabilities are evolving.
For example, Amodei said AI will allow individually innocuous data to be legally collected to provide a comprehensive picture of anyone’s life automatically and at scale.
There are other factors involved.
Amodei has been a critic of Trump, supported Kamala Harris in last year’s election, and has been a vocal advocate of regulating AI, despite Trump’s executive order reversing relatively light safety and privacy mandates that Joe Biden had imposed on the field.
In announcing the ban on Anthropic’s technology, President Trump labeled Anthropic a “radical left” company and its executives as “left wing lunatics” who have no idea what the real world is and are out of control. This regime has declared war on its opponents and critics.
Mr. Hegseth directed the department to establish benchmarks for “model objectivity” in procurement standards, “eliminating utopian idealism and embracing hard-nosed realism,” and said “diversity, equity, inclusion, and social ideology” have no place in the Army Corps.
The action taken against Anthropic is unprecedented.
The Pentagon not only stopped doing business with the company, but declared it a “supply chain risk.”
In a social media post, Hegseth wrote that no contractors, suppliers, or U.S. military partners may conduct commercial activities with Anthropic, which would prevent the company from working with partners in the incestuous AI ecosystem and could force companies like Nvidia, Amazon, and Google to sell stock and end commercial relationships with Anthropic if they want to do business with the Department of Defense.
Some AI analysts say this could be a crushing blow to companies considered to be at the forefront of AI development. Anthropic’s Claude Code tool released last month wiped out more than $1 trillion in value for software companies.
In addition to terminating its relationship with Anthropic, the administration is attempting to blackmail its business partners. President Trump also threatened to use the Cold War-era Defense Production Act to force supplies to Claude without any warning.
In other words, by declaring this a supply chain risk and ordering other companies not to address it on pain of losing their own defense contracts, the administration is saying it is a risk to national security. But at the same time, this shows that Claude is so important to national security that he must be made available without restriction.
Amodei said the company could not in good conscience comply with the administration’s demands and would challenge it in court. However, the only information the company has about the conflict so far comes from tweets from Trump and Hegseth.
He believes that at this stage of AI’s development, there are potential risks to humanity if Claude were deployed in fully autonomous mode, and that Fourth Amendment privacy protections could be undermined if AI were used to collect information about American citizens without changing existing privacy laws.
The obvious solution to the conflict would be for Anthropic to simply cease doing business with the Department of Defense, and for the Pentagon to recruit other AI companies to replace it, as it has in the past. It has already signed a deal with OpenAI and is close to a deal with Elon Musk’s xAI.
Anthropic believes that AI is too immature and dangerous to deploy without the limits it claims, and argues that current laws have not kept up with the rate at which AI capabilities are evolving.
The $200 million ($281 million) deal is insignificant for Anthropic, which just raised $30 billion at a $380 billion ($535 billion) valuation. It would be willing to go away, but the administration is not ready to allow that to happen. This shows that Claude is the only product currently available that can provide the Department of Defense with what it wants.
To his credit, Mr. Amodei is holding his ground because the issues at stake are far more important, and because this administration, which has a history of simply ignoring American laws and the Constitution, clearly has no faith in using AI tools responsibly.
He clearly believes that Claude has not been developed to the point where he is reliable enough to control autonomous weapons without human supervision.
Although Claude is known for having fewer “hallucinations” (inaccurate or fabricated responses) than his rivals, hallucinations still occur. AI is currently an imperfect technology.
Anthropic’s case highlights the inadequate protections under current law against the risks of technology malfunctions and misuse, or the risks of allowing AI to be applied before the risks are understood.
President Trump also issued an executive order rescinding a Biden executive order that established certain regulations on AI that placed more emphasis on transparency of companies’ activities than on restricting AI development, and an executive order aimed at overriding state laws deemed to limit AI development. The administration wants a “minimized burden” national framework for AI.
AI is the most important technology of our time. It threatens to change the way we work and live, the relationship between government and its people, and the way we wage war. There should be guardrails.
Instead, global competition is taking place, despite the recognition that the technology is not fully developed and carries significant risks (even closer-knit European countries are talking about rolling back ties).
Anthropic’s stance is largely supported by the people at the center of AI development – the people building the models, not the billionaires profiting from them, suggesting that the people who know the technology best are very conscious of the risks.
Rather than leaving it up to people like Trump, Hegseth, or Amodei to develop a moral framework for how AI is deployed, we desperately need lawmakers to get involved and create a 21st century legal framework for a 21st century technology that promises so many opportunities but also holds so many potential risks.
Our Business Briefing newsletter brings you top stories, exclusive coverage and expert opinion. Sign up to receive it every weekday morning..
Stephen Bartholomews is one of Australia’s most respected business journalists. He was most recently co-founder and deputy editor of the Business Spectator website and deputy editor and senior columnist for The Australian.Connect via email.
