On the evening of August 28th, a federal judge in California ruled that the Trump government had previously outlawed the listing of AI Anthropic as a “supply chain risk”. This was the first successful court in Anthropic ' s case and caused a major setback to the Pentagon ' s restrictive measures against the company.
It's a dispute that models use boundaries.
In his ruling, Judge Rita Lin stated that the United States Secretary of Defense, Pete Hegseth, had identified Anthropic as a national security risk, as “unlawful retaliation”, in violation of the First Amendment to the United States Constitution. She also considers that this decision was “arbitrary and capricious” and concludes that Anthropic did not enjoy the due process guarantees required by the Fifth Amendment.
Earlier this year, Trump and Hegseth asked federal agencies to cease their cooperation with Anthropic, not only in the defence system but also in other government departments. The controversial fuse is that Anthropic places clear security restrictions on the use of models, does not agree to their use for fully autonomous weapons, and massive surveillance against American citizens.
The court questioned the inconsistencies in the Government's reasoning.
The Pentagon stated that the model would be used only for legitimate purposes and accused Anthropic of attempting to continue to restrict the use of the military after government procurement. In his judgement, however, the judge wrote that the Government ' s statements and actions showed that the measures in question were more like publicly punishing Anthropic for criticizing the Government than based on real supply chain security risks.
She also noted that there was a clear contradiction in Government practice itself. For example, Hegseth had proposed the application of the Defence Production Act to Anthropic, which meant that the Government had once considered the company “essential” to national security rather than a threat. At the same time, the Ministry of Defence has not completely severed its links to Anthropic, is advancing its contract with the company and continues to work with its new model Mythos on cybersecurity matters.
National security reasons not supported
The judgement also refers to the fact that Anthony did not have so-called “backdoor control” after having delivered the technology to the Ministry of Defence. On this basis, the judge concluded that the Government had described it as a security threat, lacking factual support. She wrote that the Ministry of Defence was free to choose the AI supplier of cooperation, but that the extensive restrictions imposed on Anthropic were illegal and unfounded.
Anthropic filed two lawsuits against the United States Department of Defense in California and Washington, D.C., in March this year. This decision comes from the California case and proceedings in Washington, D.C., are still ongoing.