The Trump administration stood before federal judges Tuesday and insisted that Anthropic poses a national security threat to the U.S. military, even as internal deliberations are underway to license the company's most advanced AI model for defense against cyber attacks.
The contradiction did not escape the bench. During oral arguments in the D.C. Circuit Court of Appeals, Judge Karen Henderson expressed skepticism of the government's position, noting an absence of evidence supporting allegations of malicious intent. "To me this is just a spectacular overreach by the department," she said, pushing back against Pentagon claims that Anthropic's commitment to AI safety principles constitutes a reliability risk.
The core dispute turns on whether a private company can be permanently blacklisted from federal contracts because it refuses to adopt the Pentagon's operational standards. Anthropic has declined to agree to what the military calls the "all lawful use" framework, citing concerns about mass surveillance applications and autonomous weapons systems. Pentagon officials argue this refusal makes the company unreliable, pointing to what they characterize as "ideological" positions on how its technology should be deployed.
Judge Gregory Katsas challenged this reasoning from a different angle, highlighting the practical problem of evaluating AI safety policies in a field that changes rapidly. "It doesn't really matter whether we focus on what might happen with the one they're currently using or what might happen with the one that everyone knows they will need three months from now, because AI three months from now will be totally different from the AI of today," he said.
The Pentagon's lawyer, Sharon Swingle, defended the blacklist designation by framing it as a trust issue. She told the court that negotiations with Anthropic had grown increasingly hostile and that new restrictions were being imposed regularly, leaving the military with no confidence in the company's reliability. The government opted for the supply chain risk label rather than a simpler approach, Swingle argued, because time was critical.
Anthropic's legal team countered that the blacklist is a sledgehammer solution. Lawyer Kelly Dunbar noted that if the Pentagon simply distrusts Anthropic's models, it can choose not to do business with the company, a less damaging option that federal law requires agencies to consider first. The permanent disbarment not only blocks future contracts but signals to the broader commercial market that Anthropic is a national security threat, potentially undermining its business relationships across multiple sectors.
The legal split between courts in D.C. and San Francisco has created an awkward middle ground. Anthropic cannot pursue new Pentagon contracts while litigation continues, but it can maintain existing relationships with other federal agencies. The D.C. Circuit agreed to accelerate its review, potentially ruling within weeks, though complex cases often stretch longer.
President Trump gave the Pentagon until August to extract Anthropic's technology from military systems, a deadline that could force a resolution before appeals courts rule on the fundamental question of whether the company's safety commitments constitute a genuine threat or simply a philosophical disagreement with Pentagon doctrine.
Author James Rodriguez: "The government wants to have it both ways, and the judges aren't buying it. Either Anthropic is a security risk or it's not, and this simultaneous blacklisting while courting their best model exposes the weakness of the Pentagon's argument."
Comments