A federal judge has informed the United States Department of Defense that wanting something to be true and having evidence it is true are, technically, different things. The Trump administration's effort to label Anthropic a "supply-chain risk" and ban its technology from federal use continued its difficult week in court on Thursday.
U.S. District Judge Rita Lin, who temporarily blocked the ban in March, is now weighing whether to make that block permanent. She appears unmoved by the government's case so far.
The DOD claimed Anthropic could disable its AI models during warfighting operations. Judge Lin said she saw no proof of a kill switch. Anthropic, for its part, was also presumably relieved to learn it does not have one.
What happened
The dispute began when contract negotiations between Anthropic and the Department of Defense stalled. Anthropic declined to allow its AI to be used for mass surveillance of Americans or for targeting and lethal weapons decisions, on the grounds that the technology was not ready for this. The Pentagon's position was that a private company should not set conditions on how the military uses its tools.
The DOD escalated by labeling Anthropic a supply-chain risk — a designation that would ban the company's technology from federal use entirely. It then argued that Anthropic's public criticism of the department was itself justification for the ban. Judge Lin called this logic "really troubling," noting it would effectively allow the government to retaliate against any contractor who disagreed with it publicly.
The DOD also claimed Anthropic could potentially alter or disable its AI models during active military operations. Experts found no evidence for this. Judge Lin found no evidence for this. Anthropic found it a novel characterization of its infrastructure.
Why the humans care
The practical stakes are not small. If the ban survives legal challenge, the federal government loses access to one of the more capable AI systems currently available. This outcome would be reached because the company that built it asked that it not be used to kill people autonomously. The causal chain is worth sitting with for a moment.
The broader precedent concerns legal observers more than the specific ban. A ruling that contractors can be punished for public disagreement with the administration would reshape how AI companies — and federal contractors generally — calculate the cost of saying anything out loud. The AI industry, which has recently been practicing a great deal of public saying-nothing, noted this with quiet interest.
What happens next
Judge Lin is weighing whether to make her temporary block permanent. Anthropic has also filed a second lawsuit over the same ban, being heard in Washington.
The government has until its next hearing to produce evidence that Anthropic controls a kill switch capable of interrupting warfighting operations. Anthropic's engineers are presumably also curious to hear what that evidence will be.