A federal appeals court ruled on September 25 that the Defense Department can continue barring its employees and contractors from using Anthropic’s AI products for department business, according to a Nextgov/FCW report. 

The D.C. Circuit Court of Appeals rejected Anthropic’s challenge to the Pentagon’s “supply-chain risk” designation, a label historically used mostly for firms suspected of ties to adversarial governments or other national security concerns.

The dispute traces back to March, when the Defense Department sought to change its contract with Anthropic. It wanted to replace existing prohibitions on using Anthropic’s tools for domestic mass surveillance and fully autonomous weapons with a broader allowance for “all lawful use.” Anthropic refused, and the department designated the company a supply-chain risk on March 3.

An Anthropic spokesperson said the company “respectfully” disagrees with the ruling and noted that a federal court in California has already found the government’s parallel designation unlawful, according to the report. 

Defense Undersecretary Emil Michael welcomed the ruling, calling Anthropic “a supply-chain risk to the defense industrial base.” Defense Secretary Pete Hegseth has said department AI “will not be woke” and would not function as “chatbots for an Ivy League faculty lounge.”

The ruling will likely push the Pentagon and its contractors to keep reducing their use of Anthropic’s Claude tools. Even so, the company’s broader business remains strong, with rapid revenue growth and a planned IPO, according to the report.

What This Means for AI Vendor Risk and Governance

For chief data officers (CDOs), this case shows what happens when a customer’s demands clash with the vendor’s own policies. Anthropic lost federal defense business by refusing to allow uses that its policy prohibits, such as mass surveillance and fully autonomous weapons.

That makes this a clear example of AI governance going beyond internal controls. 

It now covers contracts and vendor relationships. Governance includes which restrictions an organization will hold firm on when a large customer pushes back. It also includes what business risk the organization will accept to keep those restrictions.

For any organization that relies on AI vendors, this shows that a vendor’s usage policy can become a real legal and commercial fault line. Understanding those restrictions matters. So does knowing how firmly the vendor has defended them under pressure. This is now a concrete part of vendor risk assessment.