Decisions about lawful military uses must rest solely with the Pentagon, not a private company
At the intersection of artificial intelligence, national security, and the global movement of talent, the U.S. Department of Defense has formally challenged its relationship with Anthropic, alleging that the AI company's international workforce and leadership conduct pose unacceptable risks to American military operations. The Pentagon's court filing reflects a deeper tension that has been building quietly beneath the surface of Silicon Valley's defense partnerships: who ultimately governs the tools of war when those tools are built by private hands reaching across borders? The dispute, now headed toward a March 24 hearing, may force a reckoning with how the nation balances technological openness against the imperatives of strategic security.
- The Pentagon alleges that Anthropic's Chinese employees could be legally compelled by Beijing to surrender sensitive AI intelligence, creating a potential pipeline of secrets to an adversarial power.
- DoD undersecretary Emil Michael accuses Anthropic's leadership of deliberately restricting military use of Claude, a move the government frames as actively degrading American defense capabilities.
- The filing claims Anthropic broke faith by leaking confidential Pentagon discussions to journalists, transforming a contractual dispute into an allegation of institutional betrayal.
- The Pentagon has declared the supply chain risk at a 'saturation point,' warning that disruption to Claude during live military operations could cost human lives.
- More than 30 employees from Google, OpenAI, and Google DeepMind have filed a brief supporting Anthropic, signaling that the AI industry is watching this case as a test of its own autonomy.
- With a 180-day severance clock now running and a court date imminent, the outcome threatens to redraw the boundaries between private AI development and national defense partnerships.
The Pentagon has filed a formal legal complaint against Anthropic, raising alarms about what it describes as a fundamental security vulnerability: the company's significant employment of foreign nationals, including workers from China. At the heart of the concern is Beijing's National Intelligence Law, which can compel Chinese citizens to cooperate with state intelligence efforts — a risk the DoD argues is unacceptable given the sensitive AI systems these employees help develop.
But the government's grievances extend well beyond hiring practices. Pentagon undersecretary Emil Michael's declaration accuses Anthropic's leadership of deliberately restricting how the military could use Claude, the company's flagship AI model. While other AI firms working with the Pentagon also employ international talent, Michael argued they have demonstrated trustworthy security practices — a standard he contends Anthropic has failed to meet. In the DoD's view, decisions about lawful military applications are the government's alone to make, not a private company's.
The allegations grow sharper still. The Pentagon claims Anthropic leadership leaked confidential, unclassified discussions with the DoD to journalists — an act framed as bad faith that compounded the existing concerns. Together, these issues have led the government to declare that supply chain risks have reached a breaking point, warning that any disruption to Claude during an active military operation could result in loss of life.
In response, the Pentagon announced it will discontinue all Anthropic defense products within 180 days. Neither party has commented publicly. The case has attracted notable industry attention, with more than 30 employees from Google, OpenAI, and Google DeepMind filing a brief in support of Anthropic. A court hearing set for March 24 could ultimately redefine how the U.S. government structures its partnerships with private AI companies operating in an increasingly interconnected world.
The Pentagon has formally raised alarms about what it sees as a fundamental security vulnerability at Anthropic: the company's reliance on a substantial workforce of foreign nationals, including employees from China. In a court filing, the Department of Defense laid out a stark concern—that these workers could be compelled by Beijing's National Intelligence Law to hand over sensitive information about the artificial intelligence systems they help build and maintain.
But the DoD's complaint goes well beyond the simple fact of foreign hiring. Pentagon undersecretary Emil Michael's declaration paints a picture of a company whose leadership has, in the government's view, actively obstructed legitimate military use of Anthropic's flagship AI model, Claude. While other major U.S. AI labs working with the Pentagon may employ foreign nationals too, Michael argued, those companies have demonstrated strong security practices and a track record of trustworthy collaboration. Anthropic, he contended, is different.
The core grievance centers on what the Pentagon describes as deliberate obstruction. Anthropic's leadership, according to the filing, knew full well that adversarial nations were attempting to steal the company's AI technology for unrestricted military purposes. Yet the company imposed restrictions on how the Pentagon could use Claude, restrictions that Michael's declaration says would weaken American military capabilities relative to those of hostile powers. The Pentagon's position is unambiguous: decisions about lawful military applications belong to the Department of War alone, not to a private company.
The allegations escalate further. The DoD claims Anthropic's leadership acted in bad faith by leaking unclassified information from confidential discussions with the Pentagon to journalists. This breach of trust, combined with the foreign workforce risk and the restrictions on military use, has led the Pentagon to conclude that the supply chain risks have reached what it calls a saturation point. The stakes are framed in the starkest possible terms: if Anthropic were to shut down access to Claude or alter its functionality during an active military operation, the resulting harm to national security could include loss of human life.
In response, the Pentagon has announced it will sever all use of Anthropic's defense products within 180 days. The company and the DoD have both declined to comment on the filing. Meanwhile, the legal battle has drawn support from unexpected quarters—more than 30 employees from Google, OpenAI, and Google DeepMind filed a brief backing Anthropic's position in the dispute. The next court hearing is scheduled for March 24, and the outcome could reshape how the U.S. government approaches partnerships with private AI companies, particularly those with international workforces.
Citações Notáveis
Anthropic's leadership demonstrated an intent to prevent the U.S. military's lawful use of their LLM product, Claude— Pentagon undersecretary Emil Michael, in court filing
Anthropic leadership's adversarial behavior has elevated the supply chain risks to a saturation point— Pentagon declaration
A Conversa do Hearth Outra perspectiva sobre a história
Why does the Pentagon care so much about where Anthropic's employees come from? Isn't that just normal global hiring?
It's not about hiring practices in the abstract. It's about leverage. China's National Intelligence Law can compel citizens to cooperate with intelligence gathering. If you're building military AI and your team includes people who could be legally forced to hand over what they know, that's a structural vulnerability the Pentagon sees as unacceptable.
But the filing also mentions something else—that Anthropic deliberately blocked military use of Claude. That seems like a separate issue from who works there.
It is separate, but the Pentagon is arguing they're connected. The company knew adversaries were trying to steal the technology, yet still imposed restrictions on how the military could use it. The Pentagon reads that as bad faith—as if Anthropic's leadership doesn't trust the U.S. military to use the tool responsibly.
What does Anthropic say about all this?
They haven't commented publicly. But they're fighting it in court, and they've got support from employees at Google and OpenAI. That suggests there's a real disagreement about where the line should be between a company's values and government demands.
So what happens now?
The Pentagon stops using Anthropic's products in 180 days. The court case continues. And the broader question—how much control should the government have over AI companies' decisions—remains unresolved.