S. 2750: SANDBOX Act

S. 2750, the SANDBOX Act, would create an "Artificial Intelligence Regulatory Sandbox Program" run out of the White House Office of Science and Technology Policy (OSTP). The OSTP Director would have to stand it up within a year, and the whole program would shut down 12 years after it launches. To apply, a company has to be under U.S. jurisdiction and either already be a U.S. business or become one within 180 days of signing an agreement.

The process is built to move fast. The agency that regulates the applicant gets 90 days to approve or deny, with one possible 30-day extension. Miss the deadline without filing a decision, and the Director must presume the agency has no objection. If the agency wants changes to reduce risk, the applicant has 60 days to revise and the agency gets up to 60 more to review. Approval still isn't a waiver until the Director signs a written agreement, due within 45 days.

The relief can stretch out. A waiver starts at 2 years and can be renewed up to four more times, two years each — up to a decade total. Participants would have to report incidents within 72 hours if they cause health or safety harm, economic damage, or an unfair or deceptive trade practice. The bill defines health and safety risk broadly, covering bodily harm, loss of life, or substantial adverse effects on health, including, in the bill's words, life before birth or unborn humans. Economic damage is narrower: likely tangible, physical harm to a consumer's property or assets.

The bill also turns the experiments into a deregulation pipeline. After May 1 each year, the Director must send Congress a special message naming rules that sandbox results suggest should be amended or repealed. Congress then gets an expedited track to pass a joint resolution of approval within 60 legislative days, with committees discharged if they sit on it too long. According to the bill, the rules eligible for this treatment — "covered provisions" — aren't just formal regulations; they include agency guidance, FAQs, bulletins, derivative material, and even rules Congress required by statute. Sponsors frame the program as a way to modernize outdated rules for AI; opponents, including the advocacy group Sensible Safeguards, argue it's a mechanism for routing around normal safeguards.