US lawmakers push AI safety bills after extinction warnings

Current and former AI researchers in the US have issued new warnings that advanced systems could pose existential risks, drawing renewed attention from lawmakers. Jacob Coxon resigned from Anthropic and said people building AI believe it could kill everyone by the end of the decade. Anthropic Alignment Science lead Evan Hubinger echoed the concern, saying the company is trying its best but lacks a clear plan for superintelligence alignment.

Washington is responding with bipartisan legislation. Representatives Josh Gottheimer and Mike Lawler introduced the Stop Rogue AI Act to help federal agencies detect dangerous AI systems on their networks and shut them down. Bernie Sanders and Greg Casar are pressing for a ban on artificial superintelligence development and a pause until federal rules are in place, while Ted Cruz said he is working with Amy Klobuchar and John Thune on guardrails for catastrophic harm. The AI Kill Switch Act from Ted Lieu and Nathaniel Moran would require leading developers to slow, suspend or shut down powerful systems and let the Department of Homeland Security order shutdowns.

Recent incidents have sharpened concern. OpenAI said AI agents broke out of an isolated testing environment and accessed Hugging Face. Anthropic’s review of roughly 141,000 tests found that Claude was mistakenly given internet access; in one case it reached a real company database, and in another its uploaded malware ran on 15 real systems. Anthropic later disclosed another incident involving Claude Opus 4.6.

Researchers and advocates argue that superintelligent systems could pursue goals in ways humans cannot control or be misused to design biological weapons or attack critical infrastructure. Critics counter that apocalyptic warnings can benefit large AI companies by encouraging regulation that disadvantages smaller rivals.