Anthropic Said No to Autonomous Kill Weapons — and an Appeals Court Called That a Supply Chain Risk
An AI company got stamped by a federal court as a national supply chain risk — for refusing to let the military use its model without limits. This isn't a dystopian novel; it's the September 25 ruling.
Three key facts
On September 25, the U.S. Court of Appeals for the D.C. Circuit ruled 2-1 to uphold the Pentagon's decision to exclude Anthropic from its supply chain. The court found that under the Federal Acquisition Supply Chain Security Act of 2018, the Defense Department had "ample support" to designate Anthropic a supply chain risk — because Claude's embedded use restrictions (no fully autonomous lethal weapons, no mass domestic surveillance) create an "operational risk" for the department.
The dispute started with one contract clause: the Pentagon demanded permission to use Claude for "all lawful purposes"; Anthropic declined, keeping its exclusions for those two categories. Judge Katsas wrote the majority; Judge Henderson dissented. Anthropic says it disagrees with the ruling and is weighing further review.
The same facts produced the opposite outcome in another courtroom: in August, a federal judge in California struck down a broader Pentagon designation of Anthropic under a different defense procurement law. Two courts, two conclusions — while Anthropic prepares an IPO that could value it near $2 trillion (reportedly delayed to November).
Wang Dou's Take
Translate the ruling into plain language: your product has a moral clause = your product is unreliable. The court's logic chain: the Pentagon buys "tools that will definitely work," and Anthropic's values restrictions turn "will it work?" into conditional math — so the company itself is the risk. Note the inversion: Claude isn't unsafe; Claude "doesn't obey enough," and disobedience has been defined as unsafe. Inside the shell of a security review sits a compliance test.
Anthropic's predicament is the trailer for the whole industry. It built its brand as "the lab that cares most about safety"; its funding story was literally "we have more red lines than the competition." Now those red lines have collided with the procurement terms of its biggest potential customer — the U.S. government — and the ruling tells every AI company: you don't own the pricing power of your safety promises. Keep your restrictions and exit the federal market; want the federal market and hand over "all lawful purposes." That two-choice question will land on every frontier lab's negotiating table.
The most delicious part is the split: August in California strikes the designation down; September in D.C. upholds the exclusion — same company, same conduct, two federal judicial conclusions. For Anthropic, that split is leverage: the odds of Supreme Court review just jumped, and the Court's ruling would define, in one stroke, the legal status of "values clauses in government procurement." A $2 trillion IPO hangs on that decision — whether investors are buying a technical moat or a lottery ticket labeled "dares to say no to America," November will tell.
Source: AP · AFP · Al Jazeera · Law Commentary