A group of bipartisan senators introduced new legislation Thursday that would create a dedicated artificial intelligence safety review office within the Department of Commerce, aiming to oversee and support developers of advanced AI systems.
The Preserving American Dominance in AI Act of 2024, led by Senator Mitt Romney (R-Utah), would establish an office headed by a Senate-confirmed under secretary to evaluate AI systems for potential extreme risks before deployment.
Key aspects of the proposed legislation include:
- Requiring AI developers to undergo a "take-home test" to assess their systems for safety risks
- Establishing collaboration with NIST's AI Safety Institute for technical guidance
- Mandating a 90-day review period for examining AI model safeguards
- Implementing reporting requirements focused on national security implications
- Setting penalties for non-compliance, including up to 10 years imprisonment or $1 million daily fines
The bill garnered support across party lines, with co-sponsors including Senators Jerry Moran (R-Kan.), Angus King (I-Maine), Jack Reed (D-R.I.), and Maggie Hassan (D-N.H.).
Senator Hassan emphasized the importance of protecting against potential misuse by bad actors while maintaining technological leadership: "We must continue to work to protect against potential risks to public safety, such as the misuse of this technology by terrorists or criminal organizations."
The legislation also requires Infrastructure-as-a-Service providers to implement customer verification standards for foreign transactions. Advanced data centers training frontier AI models would need to report facility locations and ownership details to the new office.
The bill aims to strike a balance between fostering AI innovation and implementing necessary safety measures to address chemical, biological, radiological, nuclear, or cyber risks associated with advanced AI development.