Pentagon Bans AI Maker Anthropic Over Its Weapons Refusal

Anthropic makes Claude, one of the leading AI assistants used by businesses and everyday people. Earlier this year, the Pentagon put the company on a supply-chain risk list — a label normally saved for untrustworthy vendors. The practical effect: military agencies and their contractors had to stop using Claude entirely by this month. On Friday, a federal appeals court in Washington, DC refused to overturn that label, ruling 2-1 that the Pentagon had solid grounds.

The fight comes down to a single contract term. Anthropic told the government it would not allow its current AI models to power autonomous weapons — machines that decide who to shoot without a human — or domestic surveillance of Americans. Defense Secretary Pete Hegseth called that refusal a significant national security risk. The judges agreed the Pentagon simply excluded a vendor that wouldn't accept a term it considered essential, not because Anthropic held unpopular political views.

Anthropic didn't walk away empty-handed everywhere. A federal judge in San Francisco threw out a separate supply-chain label in March, so one of the two designations no longer stands. But Friday's decision keeps the other one in place indefinitely, so the Pentagon's block on Claude continues. Anthropic says it is weighing appeals to a larger panel or the Supreme Court. Either way, lawyers expect years of litigation.

The real-world fallout is mostly about money and alternatives. Anthropic said it lost revenue when customers got nervous about working with what it called a government pariah, though the company still reports growing sales and is moving toward a possible stock market debut this year. Meanwhile, the Pentagon is shifting to rival models such as SpaceX's Grok, Google's Gemini and OpenAI's GPT — moves some employees at those companies have protested on ethical grounds.