U.S. Congressman Proposes AI Landmark Bill: AI Models Cannot Self-Improve Without Government Safeguards

CNBC reports that on the 28th, Democratic U.S. Representative Ro Khanna from the Silicon Valley district plans to submit the "Human Control of AI Act." Core restrictions include: AI models cannot recursively self-improve or擅自 alter core objectives, containment methods, or shutdown controls before the government establishes safety protections and regulatory approval is obtained. The bill also requires the establishment of a new regulatory body, independent audits, and the inclusion of chips in regulation. Khanna stated that the risk of civilizational extinction is real.