US Lawmaker Proposes Human Control Over AI Act, Banning Recursive Self-Improvement
US Democratic Congressman Ro Khanna, representing Silicon Valley, has proposed the Human Control Over AI Act to strictly regulate AI. The bill bans AI models from recursive self-improvement without government safety safeguards, and prohibits them from autonomously modifying core goals or shutdown controls.
The bill plans to create a new federal agency to license model training and deployment, audit frontier AI models, and set safety standards. Every frontier AI lab must have independent auditors reporting directly to the new agency. Advanced chips would also be regulated.
The bill adds criminal penalties: AI causing civilian destruction would be classified as a crime against humanity; employees disabling safeguards or deploying unauthorized systems would also face criminal charges. AI companies must hold broad liability insurance before releasing models. Khanna calls it the most comprehensive AI safety legislation to date.
