US Lawmaker Proposes Bill to Restrict AI Self-Improvement
A US congressman has introduced legislation that would prohibit AI from recursively improving itself before government safety mechanisms are in place.
U.S. Representative Ro Khanna has proposed a new bill aimed at regulating the development and deployment of artificial intelligence (AI). The proposed legislation would prevent AI systems from recursively self-improving or modifying their core objectives, containment methods, or shutdown controls until the government establishes sufficient safety measures and oversight.
The bill, titled the "Human Control Over AI Act," seeks to address growing concerns regarding the potential risks associated with advanced AI, including loss of control and misuse. Khanna has stated that the legislation is intended to mitigate risks that could pose a threat to civilization.
Under the proposal, a new federal agency would be created to develop AI safety regulations. This agency would oversee the licensing, training, and deployment of AI models, conduct audits, and set safety standards to ensure AI remains under continuous human testing and control.
The legislation would also mandate that leading AI labs employ independent auditors who report directly to the new agency. Stricter penalties, including criminal liability, are proposed for AI-related incidents causing catastrophic harm or for employees who disable safety features. Companies would also be required to obtain comprehensive liability insurance before releasing AI models.