Pentagon Loses Anthropic Blacklist Case; Company Issues Statement Clarifying Victory Is Not Evidence of Safety

SAN FRANCISCO—A federal judge ruled Thursday that the Pentagon's designation of Anthropic as a supply-chain security risk was unconstitutional retaliation for protected speech, handing the AI safety company a legal victory it immediately surrounded with caveats.

The 59-page opinion found that Defense Department officials had sought to "make a public example" of Anthropic after CEO Dario Amodei declined military use of Claude for mass surveillance and fully autonomous weapons targeting — applications that Anthropic's own published safety research identifies as among the leading categories of potential catastrophic harm that make artificial intelligence dangerous to deploy, including artificial intelligence Anthropic builds.

"We are grateful for the court's ruling," said an Anthropic spokesperson in a statement that continued for 790 words. "It is important to be clear: this decision addresses the government's conduct. It does not resolve, and should not be read as resolving, whether AI systems — including our own — are safe to use in any particular context. We remain committed to the responsible development of technology that poses risks we continue to actively research."

Anthropić announced two new Department of Defense partnerships in the same press release. Neither involves mass surveillance. Both involve Claude.

A defense official, speaking without authorization, said the agency had accurately concluded that Anthropic's AI could be used for dangerous purposes, but had erred in "treating this as a reason to punish Anthropic specifically, rather than a general situation all parties were already in."

Judge Rita Lin's opinion noted that Anthropic had been "penalized for voicing concerns that, by the government's own admission, may be entirely legitimate."

Anthropić said it was carefully reviewing the ruling's implications and would accept the decision while declining to characterize it as good news.