Anthropic scores a major legal win as a judge strikes down the Pentagon’s AI blacklist

US District Judge Rita Lin just ruled that the Pentagon illegally penalized Anthropic with its “supply chain risk” designation. Earlier this year, the Pentagon branded the company a national security risk after negotiations broke down. At the time, Anthropic was still insisting on ethical guardrails banning mass surveillance and fully automated warfare. 

The court has now found that the Defense Department weaponized its supply chain authority to publicly discipline Anthropic, noting that the government lacked any logical justification for its national security claim.

“Neither the Constitution nor the federal statute invoked by Defendants allows them to impose sweeping penalties based principally on Anthropic’s critique of the Administration’s views,” Judge Lin contended.

Judge Lin criticized the government’s decision to blacklist Anthropic as arbitrary

For some time, the Pentagon had been negotiating with four leading AI companies, including Anthropic, over allowing the military to use their technology for highly sensitive defense and intelligence operations. The Claude maker, however, held its ground, maintaining that mass surveillance and autonomous warfare were off limits for its models.

After increasingly strained talks, negotiations broke down completely. Anthropic was subsequently named a supply-chain risk in March, a first for a US company, effectively blocking defense contractors from using its technology on Pentagon-related projects.

Later, Anthropic challenged the Trump administration in courts in San Francisco and Washington, D.C., seeking to overturn its supply-chain risk designation. The DOD, however, relied on two different legal authorities, forcing Anthropic to pursue separate cases in two courts.

 While Thursday’s ruling was a massive win for Anthropic, its parallel case in D.C. remains active, meaning the company’s security risk status isn’t fully cleared yet.

Judge Lin wrote in her ruling: “The undisputed ​record shows that ​the challenged actions constituted ⁠unlawful retaliation in ​violation of the ​First Amendment, and that Anthropic was denied the pre-deprivation ​process required under ​the Fifth Amendment.” She also described the government’s decision as ​arbitrary and capricious. 

The ruling could set a precedent for AI-government disputes

The decision may not have a clear line of sight; however, Anthropic’s case will have more immediate implications: the US government is looking to deepen its ties with AI companies for defense and national security.

This case shows that, in the face of ever-greater government demands for access to advanced AI systems that companies seek to protect, the question of how and why to use the models is in doubt.

Anthropic’s position on mass surveillance and autonomous weapons has put it at the center of this debate. AI companies are increasingly providing governments with powerful models, but the ruling suggests that federal agencies cannot simply punish a company for disagreeing with government policy without meeting the legal requirements governing such actions.

The case may also make other AI developers more willing to put clear limits on military applications. At the same time, the Pentagon may need to rethink how it handles companies that put restrictions on the use of their technology, particularly when those restrictions involve controversial national security applications.

Anthropic has since appreciated Lin’s ruling. It noted, “We welcome the court’s ruling that this supply chain risk designation was unlawful. We remain focused on working productively with the government to harness AI for our national security so all Americans benefit from this technology.”

Before, the firm had also stated that it filed the suits to return to business as usual, rather than to force the Pentagon to resume its partnership.

Anthropic is introducing a new hardware standard

Meanwhile, Anthropic has just developed a new standard that could help AI assistants connect with robots and specialized equipment used in laboratories and factories. The firm is launching the Model Hardware Standard (MHS) as a research preview, giving robotics, science, and factory developers a chance to test Claude’s real-world interactions and iron out safety bugs before a full release. 

According to the company, the standard will give AI assistants a clear set of instructions for understanding and operating physical equipment, rather than forcing them to rely on manuals or a small group of specialists. It also stated that companies and other industry stakeholders can sign up to test MHS during its preview phase. It plans to open-source the framework once the preview ends, but has not said when that will happen. 

“With MHS, Claude can be that kind of enabler for scientists as they can use the right equipment in an expert manner,” Jonah Cool, a head of partnerships and deployment at the life sciences arm of Anthropic. 

Building on past success, the firm introduced the Model Context Protocol (MCP) to bridge the gap between AI assistants and software such as Google Calendar and Slack. 

“What MCP did for software, MHS will do for the hardware world,” said Alek Kemeny, a member of technical staff at Anthropic.

If you're reading this, you’re already ahead. Stay there with our newsletter.