On April 8, a three-judge panel of the U.S. Court of Appeals for the District of Columbia Circuit rejected Anthropic's request for an emergency stay. This means the U.S. Department of Defense's supply chain risk determination against Anthropic remains in effect—all Pentagon contractors are prohibited from using Claude in defense projects.
The legal battle is not over. But in this round, Anthropic lost.
How the Ban Came About
The story began late last year. The Pentagon was on the verge of signing a contract worth approximately $200 million with Anthropic. After lengthy negotiations, the deal fell through.
The core disagreement was over contract language. The Pentagon wanted broad authorization to "use Claude for all lawful purposes," while Anthropic's terms of service explicitly prohibit two categories of use:
- Fully autonomous weapons systems (such as armed drone swarms operating without human oversight)
- Mass surveillance of U.S. citizens
The Pentagon's position was that Anthropic's restrictions themselves constituted a "supply chain risk," because even if a contract were signed, it could be limited at any time. In early March, the Pentagon formally placed Anthropic on its supply chain risk blacklist, banning the use of Claude in all contractor projects.
Anthropic immediately filed a lawsuit seeking to block this determination.
What the Court Said
The panel's ruling was direct:
"On balance, the equities tip in favor of the government. On one side is the relatively limited risk of economic loss to a single private company; on the other is the question of how and through what channels to secure critical AI technology for the warfighting department during an active military conflict—a core issue of judicial administration. Accordingly, we deny Anthropic's motion for a stay pending appeal."
The court acknowledged that Anthropic "may suffer some degree of irreparable harm," but found these interests "primarily economic in nature" and insufficient when weighed against national security considerations.
Two Courts, Two Logics
This case presents an interesting legal landscape: two courts have issued contradictory rulings on the same core dispute.
Last month, a federal court in San Francisco granted Anthropic a preliminary injunction barring the Trump administration from enforcing the ban against Anthropic within non-Pentagon executive branch agencies.
This week's ruling by the D.C. Circuit Court, however, allows the Pentagon-level ban to remain in effect.
Two courts, two orders—the result is an unusual split: Anthropic is excluded from the Pentagon's procurement system but can still provide services to other government agencies.
The court has moved up the date for formal oral arguments to May 19, significantly faster than originally scheduled. This indicates that neither side wants the matter to drag on.
The Ironic Twist
In the same week the ban made headlines, Anthropic disclosed Project Glasswing—granting defensive cybersecurity access to the new Claude Mythos model for approximately 40 technology and financial companies, including AWS, Apple, Cisco, CrowdStrike, Google, JPMorgan Chase, Microsoft, and Nvidia.
Anthropic claims that Claude Mythos has discovered tens of thousands of zero-day vulnerabilities in recent weeks, including a critical security flaw in the OpenBSD system that had remained undetected for 27 years. Shut out of military security, the company's civilian security business is growing larger.
There is a deep contradiction worth considering: Anthropic imposed these usage restrictions to prevent AI from being used for autonomous lethal weapons and mass surveillance—a position that is not unusual from an AI safety perspective. But these very stances became the biggest obstacle to signing a deal with the military, cited by the Pentagon as evidence of a "threat to national security."
Using values to constrain product use, leading to being labeled a security risk by the government—this logic chain is convoluted, but it is the reality unfolding now.
Anthropic stated, "We believe the court will ultimately find these supply chain determinations unlawful."
May 19, next round.
Sources: Appeals court rejects Anthropic's bid to block Pentagon blacklisting (SiliconAngle); CocoLoop; Anthropic loses appeals court bid to temporarily block Pentagon blacklisting (CNBC); Anthropic loses bid to block Pentagon blacklisting in DC court (Axios)