Representative Ro Khanna has introduced the Human Control Over AI Act, a legislative proposal aimed at establishing strict federal oversight for artificial intelligence. The bill mandates that AI models must not perform recursive self-improvement or modify their core objectives and shutdown controls until the government implements specific safety protocols and regulatory approval is granted.
The legislation proposes the creation of a new federal agency tasked with overseeing AI safety. This body will manage a licensing system for model training and deployment, conduct audits of frontier AI models, and enforce safety standards. Laboratories will be required to employ independent auditors who report directly to this new agency, ensuring models remain under human control.
Safety requirements under the bill include the use of sandbox testing environments, physical air-gapping from the internet, and mandatory emergency kill switches. Furthermore, the agency will monitor the use of advanced chips in AI development. Companies will also be required to maintain comprehensive liability insurance before releasing any models to the public.
The bill introduces severe legal consequences for AI-related incidents. Deploying a model that causes mass harm to civilians could be classified as a crime against humanity, resulting in criminal prosecution. Employees who knowingly disable safety measures, such as logging or containment systems, or deploy unauthorized systems, will also face criminal penalties.
