U.S. congressman proposes comprehensive AI safety legislation

On September 30, 2026, it was reported that Democratic Congressman Ro Khanna, representing Silicon Valley, is set to introduce a groundbreaking AI regulation bill. The proposed legislation, titled 'Human Control Over AI Act,' would impose strict liability standards and prohibit AI systems from engaging in recursive self-improvement without government-approved safety mechanisms. Khanna emphasized the existential risks posed by AI, stating that both loss of control and potential misuse must be taken seriously. The bill outlines the creation of a new federal agency tasked with establishing AI safety regulations, licensing model training and deployment, and conducting audits on advanced AI systems. It also mandates that each leading AI laboratory employs independent auditors who report directly to the new agency. The legislation includes provisions for physical isolation of testing environments, emergency shutdown switches, and oversight of advanced chip usage. Additionally, the bill introduces criminal penalties for AI-related incidents, including 'crimes against humanity' for AI models causing mass civilian casualties. AI companies would also be required to maintain broad liability insurance. Khanna highlighted that the bill is the most comprehensive AI safety legislation to date, shaped by input from non-profit AI safety organizations rather than industry leaders.