Who Will Make the Killer Systems Anthropic Refused To Build for the Pentagon?
If Anthropic won’t build weapons that fire themselves and surveillance tools that monitor Americans at scale, the Pentagon already has a list of companies that will.
On Monday, Defense Secretary Pete Hegseth summoned Anthropic CEO Dario Amodei to what a senior Pentagon official described, with characteristic subtlety, as a “shit-or-get-off-the-pot meeting.” (Axios) The ultimatum: allow the military to use Claude for “all lawful use cases,” no restrictions, or face being declared a “supply chain risk” — the same designation the U.S. government typically reserves for companies believed to be extensions of foreign adversaries. Amodei has until Friday evening to decide.
The core of Amodei’s objection is specific and, coming from a tech CEO, almost startlingly candid. He has written that a sufficiently powerful AI could “look across billions of conversations from millions of people, gauge public sentiment, detect pockets of disloyalty forming, and stamp them out before they grow.” That’s the thing he says he doesn’t want to build. The Secretary of Defense, apparently, doesn’t want some woke vendor telling him what to do. (CNBC)
Anthropic has unusual leverage here: Claude is currently the only AI model cleared to operate inside the military’s most classified systems, and it leads the field in applications like offensive cyber operations, where exchanges need to happen at speeds no human can match. “The only reason we’re still talking to these people is we need them and we need them now,” a Defense official told Axios. “The problem for these guys is they are that good.”
But leverage cuts both ways. If Anthropic won’t play, Google, Meta, and Elon Musk’s xAI are all waiting in line — and all have signaled they’ll comply with Pentagon policy without restriction. (San Diego Union-Tribune) This is where the story gets genuinely dark.
