U.S. congressman introduces groundbreaking AI safety legislation

U.S. Representative Ro Khanna, a Democrat representing Silicon Valley, is set to introduce a landmark AI regulation bill. The legislation, dubbed the 'Human Control AI Act,' mandates that AI models cannot self-improve or autonomously rewrite their core objectives without government approval. Khanna emphasized the existential risks of AI, stating, 'The risk of civilization extinction is real.' The bill requires federal oversight and regulatory approval before companies can deploy AI models, with strict controls on self-iteration. The proposal includes the creation of a new federal agency to oversee AI safety, enforce licensing, and conduct audits. Labs must have independent auditors and adhere to a safety checklist, including sandbox testing, physical isolation, and emergency shutdown switches. High-end chips used in labs are also subject to regulation. The bill imposes severe penalties, classifying AI-caused civilian harm as a 'crime against humanity.' Employees who disable safety measures or deploy unauthorized systems face criminal charges. Companies must also obtain liability insurance before releasing models. Khanna, who consulted with non-profit AI safety groups, aims to give the safety community a voice in shaping AI policy.