A U.S. appeals court has upheld the Pentagon’s decision to exclude the artificial intelligence startup Anthropic from its supply chain, affirming a move that arose from the company’s refusal to remove certain safeguards on its Claude AI model. The ruling was issued on Friday, reinforcing the Defense Department’s cancellation of Anthropic’s military contracts and barring its technology from use by other Pentagon contractors.
The dispute began earlier this year when Anthropic declined a request from Pentagon Secretary Pete Hegseth to eliminate restrictions on the Claude model’s deployment in fully autonomous lethal weapons systems and for mass domestic surveillance within the United States. In March, the Pentagon formally designated Anthropic as a supply chain risk, prompting the severance of government ties.
The legal proceedings revealed an incident involving concerns over Claude’s usage by Palantir, a Pentagon contractor, in connection with the January 3 military operation that resulted in the capture of Venezuelan President Nicolas Maduro. According to Under Secretary of Defense Emil Michael, an Anthropic executive raised objections over Claude’s involvement, which “led to alarm” and increased doubts about the reliability and safety of the software in critical military operations. The Pentagon contended that such uncertainties could endanger the lives of U.S. service members.
Anthropic has maintained that it cannot alter its AI models once delivered to the military. However, the appeals court determined that while the company does not modify deployed models, it controls the characteristics of each new version, requiring the military to update accordingly. Judge Gregory Katsas, writing for the majority, emphasized that decisions on balancing risks related to military technology are the responsibility of the President and the Secretary of War.
The court’s decision marks a rare instance of the government applying supply chain risk designations against a domestic technology firm. Following Anthropic’s exclusion, the Pentagon swiftly engaged with competitors, including OpenAI, whose ChatGPT model secured an agreement simultaneously with Anthropic’s ban, as well as xAI, Google, and Microsoft, which signed contracts in the ensuing months.
Despite Anthropic’s central role in the AI sector and its nearing initial public offering valued at close to $1 trillion, the company’s standing at the White House has reportedly deteriorated. CEO Dario Amodei’s frequent warnings regarding AI risks have been met with skepticism, including public dismissal by former President Donald Trump.
Under Secretary Michael commented on the court ruling via social media, stating that “the hammer of justice has smashed Anthropic arguments” and asserting that warfighters would be more assured knowing no private company could influence the chain of command.
The appeals court ruling conflicts with a separate decision from a federal court in California, which last month invalidated a broader Trump-era ban against Anthropic as unlawful. However, the current Pentagon action is based on public procurement regulations and remains in force.
Anthropic responded to the ruling by reaffirming its disagreement and noting the conflicting court decisions. In a statement, the company said it was reviewing its legal options, including the possibility of further appeals.