張嘉介外流影片

A judge has blocked the Trump administration from labeling Anthropic a "supply chain risk" and cutting off all federal work with the artificial intelligence firm, an early win for Anthropic in its bitter feud with the government over AI guardrails. 

U.S. District Judge Rita Lin on Thursday ruled in favor of Anthropic, which sued the federal government earlier this month for taking actions that it called an "unprecedented and unlawful" attempt to punish the company for First Amendment-protected speech.  

Lin's ruling in the case prevents the government from enforcing its supply chain risk designation against Anthropic, a move that aimed to stop private government contractors from using the company's powerful Claude AI model. It also halts an order by President Trump for every federal agency to "IMMEDIATELY CEASE all use of Anthropic's technology."

In the ruling, she called the administration's moves "Orwellian" and said they could "cripple" the company. "At bottom, Anthropic has shown that these broad punitive measures were likely unlawful and that it is suffering irreparable harm from them," she wrote.

The dispute revolves around Anthropic's push to bar the military from using Claude for domestic surveillance or to power fully autonomous weapons. The Defense Department has said it needs to maintain the authority to use AI for "all lawful purposes," and that there are already restrictions in place against those particular uses. 

The judge wrote that her ruling does not stop the Trump administration from taking "lawful actions" that were allowed beforehand, so it is free to choose a different AI provider instead of Anthropic.

Lin stayed her order for seven days, giving the government an opportunity to appeal. 

In a statement after the ruling, a spokesperson for Anthropic said, "We're grateful to the court for moving swiftly, and pleased they agree Anthropic is likely to succeed on the merits. While this case was necessary to protect Anthropic, our customers, and our partners, our focus remains on working productively with the government to ensure all Americans benefit from safe, reliable AI."

The Justice Department and Pentagon did not immediately respond to requests for comment.

What did the Anthropic ruling say?

In an often-scathing 43-page ruling, Lin wrote that the government's moves against the company "appear designed to punish Anthropic." She said the Pentagon can choose to use whatever AI products it wants, but that the government "went further."

"The record supports an inference that Anthropic is being punished for criticizing the government's contracting position in the press," she wrote. "...Punishing Anthropic for bringing public scrutiny to the government's contracting position is classic illegal First Amendment retaliation."

She pointed to some officials' heated comments about Anthropic, including a post by Defense Secretary Pete Hegseth that called the company "sanctimonious" and said it "delivered a master class in arrogance."

The judge also took issue with the Trump administration's labeling of Anthropic a "supply chain risk," a formal designation that federal law defines as a "risk that an adversary may sabotage, maliciously introduce unwanted function, or otherwise subvert" a national security system. 

Lin wrote that the government hadn't shown why Anthropic posed that kind of risk and hadn't followed the required legal processes for determining that an entity is a supply chain risk.

"Nothing in the governing statute supports the Orwellian notion that an American company may be branded a potential adversary and saboteur of the U.S. for expressing disagreement with the government," Lin said.

She said Anthropic's due process rights were likely violated because the company didn't have an opportunity to respond to the government's moves against it. She said Mr. Trump's order for federal agencies to stop using Anthropic immediately was essentially a form of "debarment," or a ban on a company contracting with the government — but usually, firms that face debarment have the ability to oppose that measure.

And she called the government's actions "arbitrary and capricious," pointing to cordial contract negotiation emails between Pentagon Chief Technology Officer Emil Michael and Anthropic CEO Dario Amodei even as the military called Anthropic a serious threat.

After the administration took action against Anthropic, Lin noted, federal agencies aside from the Pentagon quickly terminated their use of Claude, endangering its lucrative public sector business. And Anthropic has said some government contractors are worried that they could run afoul of the president's order if they use Claude, wrote Lin.

"One of the amicus briefs described these measures as 'attempted corporate murder,'" Lin wrote. "They might not be murder, but the evidence shows that they would cripple Anthropic."

Lin also formally rejected a social media post by Hegseth that said military contractors must cut off all "commercial activity" with Anthropic — which she said seemed to illegally require companies to stop using Claude on non-military work.

During a hearing in San Francisco earlier this week, Justice Department attorney Eric Hamilton conceded that a supply chain risk designation would only stop government contractors from using Anthropic's technology for military-related work, not their other business. Anthropic argued that Hegseth's post still caused damage to the company.

The roots of the Anthropic-Pentagon feud

The dispute between Anthropic and the Pentagon highlights a broader debate over how to deal with the potential risks posed by AI.

Anthropic has long been vocal about the possible dangers of unconstrained AI, and has called for governments to enact safety and transparency rules. Meanwhile, the Trump administration has argued that strict AI regulations could stifle innovation, and has accused some AI models of being ideologically skewed or "woke." 

The recent feud revolves around a set of mass surveillance and autonomous weapon-related "red lines" set by Anthropic, the only company whose AI model was deployed on the military's classified systems. The showdown comes as the U.S. military uses Claude in its war with Iran.

Anthropic has said it isn't looking to second-guess the military's decisions. But it argues that without guardrails to block AI-powered mass surveillance on Americans or weapons that can strike without human input, there's a risk of Claude making fatal mistakes or operating in a way that clashes with democratic values.

Amodei told CBS News in a late February interview: "I think we are a good judge of what our models can do reliably and what they cannot do reliably."

The Pentagon has balked at Anthropic's push for guardrails. The military says mass surveillance of Americans and fully autonomous weapons are already barred by federal law and internal Pentagon policies, respectively. 

"But we do have to be prepared for the future," Michael said in a CBS News interview last month. "So we'll never say that we're not going to be able to defend ourselves in writing to a company."

As talks between the two sides broke down last month, administration officials publicly lashed out at Anthropic, accusing the company of trying to police the military and impose its own values onto the government. Michael said Amodei has a "God-complex," and Mr. Trump called Anthropic a "radical left, woke company."

Last month, Mr. Trump ordered federal agencies to stop using Anthropic, though he gave the military six months to phase out the service, and Hegseth said Anthropic would be labeled a supply chain risk. Anthropic quickly sued.

Lawyers for the two sides faced off in person during this week's hearing in San Francisco federal court.

The Justice Department's lawyer, Hamilton, argued that labeling Anthropic a supply chain risk was warranted because the tense negotiations between the Pentagon and Anthropic had made the military fear that the company could "manipulate" its software or install a "kill switch." He said the designation was based on a "risk of future sabotage."

Lin appeared unconvinced, and said the government appeared to be saying that a company can be designated a supply chain risk because it is "stubborn" and "asks annoying questions." 

Anthropic's lawyer, Michael Mongan, argued that if Anthropic posed such a serious risk, it doesn't make sense that the government appeared open to striking a deal until the very end.

"A saboteur is not going to get into a public spat," Mongan said. "They're just going to accept the contractual term proposed by the government and then go and do ... nefarious things."