A federal judge in California, USA, ruled on the evening of August 28 local time that the Trump administration's previous designation of AI company as a "supply chain risk" was illegal. This is the first court victory for Anthropic in this matter, and it also represents a major setback for the restrictions imposed by the Pentagon against that company.
The dispute arose from the boundaries of model usage.
Judge Rita Lin stated in her ruling that the U.S. Secretary of Defense Pete Hegseth identified Anthropic as a national security risk, which constitutes 'illegal retaliation' and violates the First Amendment of the U.S. Constitution. She also deemed this decision to be 'arbitrary and capricious', and determined that Anthropic did not receive the due process guarantees required by the Fifth Amendment.
Earlier this year, Trump and Hegseth demanded that federal agencies cease cooperation with Anthropic, a requirement that was not limited to the defense sector but also extended to other government departments. The dispute was sparked by Anthropic setting clear security restrictions on the use of their models, opposing their application in fully autonomous weapons as well as large-scale surveillance of American citizens.
The court questions the government's inconsistent reasons
The Pentagon stated that the relevant models would only be used for legitimate purposes and accused Anthropic of attempting to continue to restrict the military's use of them after government procurement. However, in the judgment, the judge wrote that the government's actions indicated that these measures were more akin to a public punishment for Anthropic's criticism of the government, rather than being based on real supply chain security risks.
She also pointed out that there are clear contradictions within the government's own practices. For example, Hegseth once proposed applying the National Defense Production Act to Anthropic, which implies that the government once considered the company to be "indispensable" for national security, rather than a threat. At the same time, the Department of Defense did not completely cut off ties with Anthropic, continued to advance contracts with the company, and continued to cooperate with its new model Mythos in matters of cybersecurity.
Not supported on national security grounds.
The judgment also mentioned that after delivering the technology to the Department of Defense, Anthropic did not possess what is known as “backdoor control.” Based on this, the judge concluded that the government’s characterization of it as a security threat lacked factual support. She wrote that the Department of Defense had the freedom to choose any AI supplier for cooperation, but the extensive restrictions imposed on Anthropic were both illegal and unfounded.
Anthropic In March of this year, two lawsuits were filed against the U.S. Department of Defense in California and Washington, D.C. respectively. This ruling comes from the California case; the lawsuit in Washington, D.C. is still ongoing.
