SANDBOX Act

This bill would create a national “regulatory sandbox” for artificial intelligence. Within one year, the White House science office must set up a program that lets companies test AI products or methods on a limited basis, with temporary waivers from certain federal rules. The goal is to speed up useful AI while protecting people from harm. To get a waiver, applicants must explain the benefits, the risks (like safety, economic harm, or scams), and how they will reduce those risks. Agencies then review and can approve all or part of a request. Approved tests last up to two years and can be renewed, but can also be revoked if terms aren’t followed. Consumer rights to sue are not waived, and companies are not immune from laws outside the waiver. Companies must report serious incidents within 72 hours and clearly disclose to the public that their AI is being tested and may not work as intended. The program would end 12 years after it starts, and the government must report results each year and coordinate with states that run similar programs .

Key points:

• Who is affected: AI developers and businesses seeking to test new tools; federal agencies that enforce rules; consumers who may try early versions of AI services .
• What changes: Temporary, limited waivers from certain rules to test AI, with safety checks, required disclosures, incident reporting, recordkeeping, and no waiver of consumer legal rights .
• When: Program set up within 1 year; each waiver lasts 2 years and may be renewed up to four times; the whole program sunsets 12 years after it begins .