A federal court in the United States has deemed the Pentagon's actions against Anthropic unlawful, following disputes over the usage terms of the AI model, Claude. Judge Rita Lin nullified the designation of "supply chain risk" and suspended the resulting restrictions, as reported by Reuters.
The judge concluded that the dispute was not about the technical reliability of the chatbot or the company's ability to fulfill government contracts, but rather about its public stance on acceptable AI use cases. The court viewed this as a retaliatory measure due to these views, which violates the First Amendment of the Constitution.
Moreover, Anthropic was not given the chance to review or contest the Pentagon's allegations beforehand. According to Lin, this lack of notice infringed upon the due process guarantees established by the Fifth Amendment.
As a result, the Pentagon's decision has been annulled and agencies are prohibited from implementing related measures. The ruling took effect immediately but may be appealed by Donald Trump's administration.
Background
The conflict between Anthropic and the Pentagon arose from the military's usage conditions for the company's models. Anthropic allowed the use of Claude for national security purposes but refused to lift restrictions on mass surveillance of U.S. citizens and fully autonomous weapon systems.
Back in February, the agency demanded the removal of all restrictions for "any lawful purposes" and threatened to blacklist the firm as a supply chain threat. Following CEO Dario Amodei's refusal, the Pentagon acted on its threat.
Shortly thereafter, Trump ordered federal agencies to cease using the company's technologies within six months.
This conflict also impacted other AI developers. Following OpenAI's agreement with the Pentagon, Sam Altman urged the military not to blacklist Anthropic and later admitted that certain aspects of his own deal with the agency were mistakes.
Despite the tensions, some U.S. entities continued to use Anthropic's technologies. In April, the National Security Agency utilized the Mythos model.
Additionally, in early August, reports surfaced about a meeting involving Meta, Anthropic, Google, and OpenAI with the Trump administration, where they discussed a voluntary government safety testing program for the most powerful AI versions.