Anthropic and DOD dispute: You have to be able to afford your stance
In the dispute with the Department of Defense, Anthropic has so far afforded its stance. It is questionable if this will remain.
(Image: Michele Ursi/Shutterstock.com/Bearbeitet von heise Medien)
The dispute between Anthropic and the US Department of Defense (DOD) is marked by duplicity and intrigue.
In the last week of February, US Secretary of War Pete Hegseth demanded that Anthropic deactivate the safeguards on its large language models (LLMs) for military purposes or supply the DOD with unrestricted models. These guardrails are intended, among other things, to prevent their use in systems for autonomous killing and for the mass surveillance of US citizens. Anthropic refused. The US government reflexively began to complain and threatened, among other things, the Defense Production Act, a law that allows the government to prioritize the production of war-critical goods.
Its stance apparently comes at a price the LLM provider can afford: The pilot project with the DOD was supposed to have a financial cap of 200 million US dollars, while the company's extrapolated annual revenue amounts to 14 billion US dollars. Anthropic also received backing from OpenAI CEO Sam Altman, who praised the steadfastness on a television show.
In typical Trump fashion, an angry government statement followed in tweet form. The US President called on all government agencies to remove Anthropic models. Deadline: six months. He also declared the AI provider a supply chain risk. This designation had previously been reserved for foreign companies like Huawei.
Videos by heise
OpenAI and xAI are ready
Lost in this drama is a study by King's College in London. The researchers showed that LLMs resorted to nuclear weapons as a means of escalation in 95 percent of conflict simulations. The study, combined with Anthropic's manufacturer statements that its technology is not ready for autonomous weapon systems, should be enough for common sense. However, that has been sorely missed in the US government since Trump II at the latest.
But fortunately for Trump, there are still providers who will do anything. While OpenAI boss Altman praised Anthropic's stance on CNBC, he secretly closed a deal with the DOD – essentially on the same terms as Anthropic. This is because, according to publicly available information, mass surveillance of US citizens and autonomous killing are also excluded here. This was accompanied by a small donation of 25 million US dollars from OpenAI co-founder Greg Brockman to Trump's lobbying group. Altman responded to the subsequent criticism with the usual crocodile tears and a zigzag course regarding the alleged contract details.
In addition, Grok, the LLM from Elon Musk's company xAI, is reportedly to be used for the important tasks of mass surveillance and autonomous killing. The tool is so unregulated that it could easily generate depictions of child sexual abuse. So if models that are chronic liars and those for casual depictions of abuse have influence on the US military networks and war systems in the future, nothing can go wrong.
Anthropic is the new Huawei
In the short term, Anthropic's marketing, which nobly defended the company's ethical values, won out. In the following days, Claude overtook competitors ChatGPT and Gemini in the app stores. However, it is uncertain how the classification as a supply chain risk will affect things. Will all companies that have contracts with US defense facilities now have to completely ban Anthropic from their supply chains?
Microsoft apparently doesn't see it that way. A spokesperson told Business Insider that legal review had shown that they could continue to offer Anthropic products to their customers through their own platforms. The only exception was the Department of Defense. For projects outside the defense sector, Microsoft could cooperate with Anthropic. As Reuters writes, lawyers doubted anyway that the Trump administration has the necessary authority to prohibit the use of Anthropic by defense contractors, for example.
Of course, this doesn't mean that companies like Lockheed Martin won't still follow Trump's social media outburst and wield the ban hammer against Anthropic. But the provider's models are currently considered the absolute market leaders, even though Claude Opus still incurs the highest operating expenses. And Anthropic's LLMs appear to be central to US war efforts in Iran. So it remains to be seen whether the declaration as a national risk will actually affect Anthropic's business.
This commentary is the editorial of iX 04/2026, appearing on March 20, 2026.
(pst)
