A federal appeals court is scrutinizing the Pentagon's controversial classification of Anthropic as a supply-chain risk to U.S. national security. This designation has led to a ban on the use of the company's AI technology within government operations and has sparked a legal battle that could redefine the relationship between AI firms and federal agencies. At a hearing in Washington, two judges on the appellate panel expressed concerns about the Pentagon's characterization, suggesting it may lack a solid legal foundation.
The conflict began over the military's use of Anthropic's Claude AI chatbot. Defense Secretary Pete Hegseth's declaration in March, which labeled Anthropic as a potential risk, arose amid disputes about how the chatbot would be utilized. Judges Gregory G. Katsas and Neomi Rao, both appointed by former President Trump, questioned Anthropic’s legal counsel regarding the legitimacy of Hegseth’s actions. They asked whether the court could intervene in the Defense Department's assessment of supply chain risks, given the unpredictable nature of AI tools. “It might not work in ways the government wants it to work,” Katsas remarked, highlighting the uncertainties associated with AI technologies.
Rao supported this perspective, referring to the AI model as a “fundamental black box.” In contrast, the third judge on the panel, Karen LeCraft Henderson, expressed skepticism about the government's rationale, labeling Hegseth's decision “a spectacular overreach.” She noted a lack of evidence for any malicious intent from Anthropic, raising concerns about the potential damage to the company's reputation and revenue from such a classification.
The Pentagon's supply-risk designation has heightened tensions between the government and Anthropic, which is seeking clarity on how its technology can be used. The company has requested assurances that its AI systems would not be employed for mass surveillance or in developing autonomous weaponry. However, the government has resisted any limitations, citing compelling national security interests. In a court filing, Hegseth claimed that Anthropic could not be trusted and warned of the potential for the company to manipulate its AI models in ways that could harm military operations.
In response, Anthropic has launched a dual-front legal challenge. On March 9, it petitioned the appeals court to review the Pentagon’s declaration, calling it “arbitrary, capricious and an abuse of discretion.” At the same time, the company filed a separate complaint in federal court in San Francisco, which temporarily halted the government's actions while the case is ongoing. A judge in that case issued an injunction against the government, although the Trump administration is appealing this ruling.
During the recent hearing, U.S. Justice Department attorney Sharon Swingle expressed concerns that Anthropic was effectively asking for an “operational veto” over lawful uses of its technology. She warned there could be undisclosed restrictions within the software that the government would not be aware of. However, Anthropic’s attorney, Kelly Dunbar, countered that the company cannot modify its AI tools without explicit military consent and does not seek to influence military operational decisions. Dunbar argued that if Anthropic's technology did not meet government requirements, the Pentagon could simply choose not to purchase it instead of imposing a damaging risk designation.
Anthropic has positioned its challenges as not only a fight for its own interests but also as a broader issue affecting all federal contractors who may face similar treatment for opposing government perspectives. The outcome of this case, Anthropic v. U.S. Department of War, could establish a precedent for how AI companies interact with federal authorities and address national security concerns. As the court prepares to make a decision, the implications of their ruling will be closely watched by both the tech industry and government entities.
