Anthropic Beats the Pentagon in Court After Refusing to Let Claude Cross Two AI Red Lines

Anthropic has won a major legal battle against the Trump administration after a federal judge ruled that the government illegally retaliated against the AI company for refusing to remove restrictions on how the military could use Claude. The dispute centered on two uses Anthropic wouldn’t approve — mass surveillance of Americans and fully autonomous lethal weapons.

U.S. District Judge Rita Lin ruled that the Pentagon’s decision to label Anthropic a national security supply-chain risk violated the company’s First Amendment rights, denied it due process under the Fifth Amendment and was “arbitrary and capricious.” 

Anthropic Refused to Give the Pentagon Unlimited Access

The conflict began when the Pentagon wanted Anthropic to permit Claude to be used for “all lawful uses.”

Anthropic agreed to remove most of its restrictions but refused to budge on two major safeguards:

• Claude couldn’t be used for mass domestic surveillance of Americans
• Claude couldn’t be used to power fully autonomous weapons that select and attack targets without meaningful human control

Anthropic CEO Dario Amodei maintained that current AI systems aren’t reliable enough for those applications.

The Pentagon took a different position, arguing that a private technology company shouldn’t determine which lawful military operations the government can perform using technology it purchases.

The disagreement eventually escalated far beyond a normal contract dispute.

Trump and Hegseth Blacklisted Anthropic

On February 27, President Donald Trump ordered federal agencies to stop using Anthropic’s technology.

Defense Secretary Pete Hegseth followed by designating Anthropic a supply-chain risk and restricting military contractors from doing business with the company. 

Anthropic sued in March, arguing that the administration was retaliating against the company because it publicly criticized the Pentagon’s position.

Judge Lin agreed.

Her 59-page ruling concluded that the government’s actions amounted to unlawful retaliation protected by the First Amendment. She also found that Anthropic wasn’t given the process required before being designated a security threat. 

The Government’s Security Argument Fell Apart

One of the most damaging developments for the government involved its original justification for labeling Anthropic a security risk.

The Pentagon had raised concerns that Anthropic could potentially access deployed Claude models and disable or alter them during military operations.

But the government eventually conceded that Anthropic doesn’t have backdoor access to models once they’re deployed inside government systems and that Claude wasn’t inherently riskier than comparable AI systems. 

The remaining concern largely centered around whether the Pentagon could trust Anthropic after the company’s public criticism.

Lin found that wasn’t enough to justify treating the American AI company as a national security supply-chain threat.

She wrote that invoking national security doesn’t provide the government with unlimited authority to punish its critics. 

Anthropic Gets Its Blacklisting Removed

The ruling gives Anthropic several major victories.

Lin:

Vacated the supply-chain risk designation, struck down Hegseth’s boycott order, issued a permanent injunction against enforcement and rejected the government’s request to delay the ruling for seven days.

Anthropic told the court that leaving the restrictions in place could have reduced its defense-related revenue by 50% to 100% and cost the company billions of dollars in 2026 revenue. 

However, Anthropic didn’t win every argument.

Lin rejected the company’s claim that Trump’s directive exceeded presidential authority, and the Pentagon remains free to choose another AI provider instead of Claude. 

A separate Anthropic challenge involving another federal supply-chain designation also remains pending before the D.C. Circuit Court of Appeals. 

The Bigger Fight Is Over Who Controls Military AI

The case raises a question that will become increasingly important as companies such as Anthropic, OpenAI, Google and xAI provide advanced models to the U.S. government.

Who ultimately decides what an AI system is allowed to do — the company that created it or the government that purchased access to it?

The Pentagon’s argument is straightforward: elected governments, not private technology companies, should determine lawful military policy.

Anthropic’s position is that AI developers still have a responsibility to establish safety boundaries when current laws haven’t caught up with what the technology can do.

The court didn’t decide which side should control military AI.

Instead, it drew a different line.

The Pentagon can refuse to use Claude because it doesn’t like Anthropic’s restrictions. What it can’t do is use national security powers to punish the company for publicly defending those restrictions.

That makes Anthropic’s victory about more than Claude. It could become an important precedent as AI companies increasingly have to decide where their own red lines sit when governments want to deploy their most powerful models.