Judge Says Pentagon Was Wrong to Punish AI Company Anthropic Over Safety Limits

On August 27, 2026, a federal judge in California ruled that the Trump administration's punishment of the AI company Anthropic was illegal. U.S. District Judge Rita Lin said Defense Secretary Pete Hegseth had broken the law by labeling Anthropic a risk to national security. The judge found that the label was retaliation against Anthropic for exercising its free speech rights under the First Amendment. She also said the decision was made without a fair process, violating Anthropic's rights under the Fifth Amendment (TechCrunch).

The label in question is called a "supply chain risk" designation. Think of it like a blacklist. When the government puts a company on this list, federal agencies are told they cannot do business with it. This tool has traditionally been used to block companies that might be connected to foreign spies or that could compromise critical infrastructure. In this case, it was turned on a domestic AI company for an entirely different reason.

The conflict started in early 2026. Hegseth and President Donald Trump labeled Anthropic a supply chain risk and ordered every federal agency, not just defense agencies, to stop working with the company. The reason: Anthropic had set firm safety limits on how its AI models could be used. Specifically, the company refused to allow the Pentagon to use its models for fully autonomous weapons and for mass surveillance of American citizens.

Hegseth announced the directive on February 27, 2026, via X, directing what is now called the Department of War to designate the company a supply chain risk (Anthropic). Days later, on March 5, Anthropic received a letter from the department confirming the designation (Anthropic).

Anthropic fought back by filing two lawsuits against the Department of Defense in March 2026, one in California and one in Washington, D.C. The company called the government's actions an "unlawful campaign of retaliation" (NPR). The D.C. case was still ongoing as of this week's ruling.

The California case moved through several stages. A federal judge temporarily blocked the Pentagon's designation on March 27, 2026 (ABC7 News). The Pentagon appealed, and a preliminary injunction was put on hold for seven days while the appeal proceeded (Inside Defense). By July 30, 2026, Judge Lin indicated she was likely to permanently block the designation, which led to this week's final ruling (Courthouse News).

One detail in the judge's reasoning stood out. She noted that the Department of Defense kept pursuing a contract with Anthropic even after slapping the company with the risk label. The department was also still collaborating with Anthropic on a cybersecurity model called Mythos.

That contradiction mattered. The government was trying to keep Anthropic out of the federal ecosystem entirely, while at the same time working with the company on a defense-related project. The judge appears to have used this inconsistency to support her finding that the designation was arbitrary and capricious.

The broader context here is about how far the government can go when it disagrees with a contractor's policies. Supply chain risk labels were designed to protect against genuine national security threats from foreign actors. Using that same tool against a domestic company because it refused to lift safety guardrails on military and surveillance applications is a major expansion of what the label was meant to do.

The case is not fully over. The Pentagon has already appealed once, and the separate lawsuit in Washington, D.C. is still making its way through the courts. That suggests the executive branch intends to keep fighting. For technology companies that work with the government, the immediate result is that the designation is gone. What happens next depends on how higher courts rule on the government's authority to use these labels in disputes over how AI is deployed.

If AI companies can set firm limits on autonomous weapons and domestic surveillance without being shut out of all federal contracts, the rules for negotiating with the government become clearer. That is good for the sector, even with the remaining legal uncertainty.