Anthropic in "legal limbo": two US courts cannot agree on the military's right to use Claude

Anthropic finds itself in a difficult legal position: a U.S. appeals court has issued a decision that directly contradicts a ruling from a lower court issued in March.

At the center of the disagreement is the question of whether the American military can use the Claude model, and if so, to what extent. Until the conflict between courts is resolved, the company is effectively stuck in a legal zone of uncertainty, which it describes itself as a "supply chain risk." The essence of the legal conflict lies in the mismatch between two judicial rulings.

The lower court, which issued its ruling in March, took a specific position on how and under what conditions Anthropic is permitted to license its technologies to federal structures. The appeals court holds a different view. When two courts diverge on the same legal question, neither party—neither the developer nor the client—can confidently rely on an agreement that may be revised by a higher court.

This is precisely the situation Anthropic now faces. For the company, this is particularly sensitive given its ambitions. Claude occupies one of the leading positions among language models, competing with OpenAI's GPT-4o and Google's Gemini.

Anthropic actively promoted its development to government and corporate sectors: Amazon became a key ally in this effort, investing billions of dollars in the company and integrating Claude into the AWS cloud platform. Any legal shadow on the possibility of working with military clients directly impacts revenue and negotiating positions. Anthropic's valuation runs into tens of billions of dollars—and a significant portion of that figure is justified precisely by prospects for growth in the government sector.

The term "supply-chain risk," which Anthropic itself uses, points to a broader problem than appears at first glance. In the world of software and AI, the "supply chain" is a complex structure of dependencies: a base model, API, cloud provider, partnership integrations, end applications. If the legal status of the base model itself is in question, instability spreads throughout the chain—from integrator startups to major defense contractors who have already embedded Claude in their workflows.

Over the past two years, American military and intelligence services have progressed from cautious experiments with AI to full-scale deployment of language models in intelligence operations, logistics, and strategic planning. Competition for defense AI contracts is being waged between the largest market players—Microsoft, Google, Oracle, Palantir. Anthropic, positioning itself as a company focused on safe and predictable AI, seemed like a particularly attractive partner for organizations that cannot afford unpredictable systems.

Now military customers have legal grounds to pause and reconsider the degree of their dependence on Claude. Conflicts between courts are typically resolved through a higher court—either by bringing the case to the Supreme Court or by establishing new precedent in the appellate circuit. Until that moment, Anthropic is forced to operate in a state of uncertainty: signing contracts knowing that their legal foundation may be revised.

For the entire industry, this situation is a telling test case. OpenAI, Google, and other contenders for government AI contracts are watching how events unfold: the legal framework that courts ultimately establish will either open doors for widespread AI implementation in the defense sector, or create restrictions that the entire industry will have to account for.