US senators unveil 'responsible' defense AI oversight bill

The Coons–Reed bill is a pivotal marker in how quickly national security institutions are trying to domesticate frontier AI. Rather than banning military AI outright, it codifies a set of guardrails: humans must stay in the loop, autonomous systems must have manual overrides, and AI can’t be the one to green‑light nuclear launches or conduct mass surveillance on US soil. That’s a concrete attempt to carve out a “responsible use” lane for increasingly capable models in defense. For the AGI race, this is less about slowing capability and more about shaping where and how it’s deployed. The bill is a direct response to the ugly fight between the Pentagon and Anthropic over classified use of Claude, which ended with Anthropic being labeled a “supply chain risk.” It signals that Washington expects frontier labs to play by a defined, legally enforceable framework if they want government contracts. Over time, similar constraints are likely to trickle into export controls, alliance agreements, and procurement standards across NATO and beyond.