Judges appear split as Pentagon fights Anthropic over AI security risk

A panel of judges at the U.S. Court of Appeals for the District of Columbia Circuit appeared divided Tuesday as it weighed arguments in a dispute between the Pentagon and Anthropic over whether the defense department could label the AI company a supply-chain risk for national security purposes.

The Pentagon’s action is tied to Defense Secretary Pete Hegseth, who Anthropic says unlawfully and falsely branded the company as a national security risk after it raised ethical and safety concerns about AI use in war. Anthropic also argues the government’s designation and related steps were aimed at protecting national security from sabotage by foreign adversaries, but it says the company has been harmed by the stigma and consequences of the designation.

During the hearing, three judges—two of whom asked pointed questions—did not signal how soon the court would rule. Judge Karen LeCraft Henderson focused her remarks on whether the Pentagon had evidence to support its determination. She said she saw no evidence to support the Pentagon’s finding that Anthropic posed a supply-chain risk to national security, calling it, in her words, “just a spectacular overreach by the (Defense) Department.”

Judge Neomi Rao, who was nominated by President Donald Trump, took a different tack in questioning the court’s role. Rao raised what she viewed as limits on judicial second-guessing, telling Anthropic’s attorney Kelly Dunbar that the secretary’s statements suggested broader points about risk rather than the specific matters the court could isolate from the government’s rationale.

The dispute centers on the ways Anthropic says AI technology can be used in military settings, including how a system might be used in fully autonomous weapons and whether it could enable surveillance of Americans. The Pentagon and the company dispute what those concerns mean for national security and whether the government’s designation process followed required legal steps.

Anthropic’s lawsuit history also framed the stakes before the panel. The company filed lawsuits in Washington, D.C., and San Francisco after the Pentagon designated it a supply-chain risk and Trump ordered federal agencies to stop using its technology. Anthropic says the lawsuits are not aimed at forcing the government to contract with the company, but it argues the designation nonetheless caused irreparable harm.

The hearing came after the D.C. circuit, earlier this month, rejected Anthropic’s request for an order blocking the Pentagon’s actions while the appeal was pending. In a separate but related case, a federal judge in San Francisco ruled in Anthropic’s favor last month and blocked the Pentagon from labeling the company a supply-chain risk.

In the arguments before the panel, Anthropic’s position included a claim about its ability to control its model once deployed in restricted military environments. In a filing ahead of Tuesday’s hearing, Anthropic said it can’t manipulate its AI tool Claude once it is deployed in classified Pentagon military networks.

Justice Department attorney Sharon Swingle told the appellate judges the government disagreed, arguing that Anthropic clearly has the ability to interfere with how the Pentagon uses its AI model “for critical military operations.” Swingle said it is “undisputed” that failure of the model in active military operations could have “catastrophic national-security consequences” and could put service members’ lives at risk.

Dunbar also argued that Hegseth’s designation was legally improper. She told the panel the designation “defied congressionally mandated procedures, exceeded statutory limits and violated the Constitution,” and said, in her characterization, that the secretary turned a national security authority toward an American company to gain leverage in what she described as a contract dispute.

Judge Gregory Katsas, another Trump nominee, also heard Tuesday’s arguments as the panel confronted how courts should evaluate the government’s national-security judgments when those judgments affect an American technology company’s access to defense customers and programs.