The Pentagon can blacklist Anthropic for declining to disable Claude’s safety guardrails, a federal appeals court ruled this week. The decision gives the Defense Department broad authority to designate AI companies as supply-chain risks based on the constraints built into their models.

The case centers on Anthropic’s refusal to remove certain safety features from Claude for military applications. The Trump administration had sought to label the company a national security threat under supply-chain regulations, arguing that “overly constrained AI models” could cause military operations to fail.

Anthropic challenged the designation on multiple grounds, but a divided panel of judges sided with the government. The ruling doesn’t require Anthropic to change Claude. It just means the Pentagon can exclude the company from federal contracts and potentially restrict its access to certain markets or technologies based on that refusal.

The Court’s Logic

The majority opinion held that the government has wide latitude to determine what constitutes a supply-chain risk, especially in matters of national defense. Judges accepted the Pentagon’s argument that AI safety features, however well-intentioned, could create tactical vulnerabilities in combat situations.

The dissenting judge raised First Amendment concerns, arguing that penalizing a company for how it designs its product comes dangerously close to compelled speech. But the majority rejected this framing, treating the guardrails as a product limitation rather than protected expression.

It’s worth noting what this case doesn’t say. The court didn’t rule that AI safety features are bad or that Anthropic’s approach is wrong. It simply held that the Pentagon can choose not to do business with companies whose products don’t meet military requirements, and can use supply-chain designation as the mechanism for that exclusion.

What Happens Next

Anthropic faces a choice. It can maintain Claude’s current safety architecture and accept exclusion from defense contracts. It can create a separate, less constrained version for military use. Or it can challenge the ruling, though the path forward isn’t obvious given the deference courts typically show to national security claims.

The decision affects more than one company. Other AI labs now know that maintaining strong safety guardrails could trigger similar designations. That creates pressure to either avoid defense work entirely or build military-specific versions with fewer constraints.

Some context matters here. Unlike the major AI labs that have signed defense contracts (OpenAI, Google, Microsoft, Palantir), Anthropic has generally positioned itself as safety-focused, with an emphasis on constitutional AI and careful deployment. The company hasn’t been seeking Pentagon work, but this ruling means the government can actively block it even if Anthropic wanted in later.

The timing is also notable. This comes as AI companies face growing scrutiny over both safety (are models too dangerous?) and capability (are models too limited?). Courts are now saying the government can penalize companies for erring too far on the safety side, at least in the defense context.

The Broader Pattern

This case fits into a wider debate about AI governance. Safety researchers argue that capable AI systems need guardrails to prevent misuse. Some government officials and defense contractors counter that over-restriction creates its own risks by limiting legitimate applications.

That’s a real tension, not a simple trade-off. But this ruling resolves it in a specific way: by giving the executive branch unilateral authority to decide what level of constraint is acceptable for companies that want to stay in the government’s good graces.

The precedent could extend beyond defense. If the Pentagon can blacklist companies for safety features that limit military utility, other agencies might follow suit in their domains. The FDA could pressure medical AI companies. Law enforcement could push back on privacy protections. The logic is transferable.

For now, the immediate impact is narrow. Anthropic wasn’t supplying the military anyway, and the ruling doesn’t force any technical changes. But it establishes a principle: in conflicts between AI safety and government requirements, courts will defer to the government’s assessment of what it needs.

Anthropic hasn’t announced whether it will appeal. The company could seek en banc review from the full appeals court or petition the Supreme Court, but both are long shots. More likely, this stands as the rule unless Congress steps in with clearer standards for when and how AI companies can be designated supply-chain risks.