US Rep Ro Khanna proposes the Human Control AI Act, banning recursive self-improvement and autonomous objective rewriting without government approval. The bill establishes a new federal agency to license models, conduct independent audits, and regulate high-end chips. It mandates strict safety measures including sandbox testing and emergency shutdowns. Violations causing civilian harm are classified as crimes against humanity, with employees facing criminal charges for bypassing safeguards. Khanna emphasizes this legislation prioritizes safety community input over industry interests to mitigate existential risks.