- The memo recommends that AI governance be organized by sector and outcomes, with regulatory attention focused on high-consequence use cases that can significantly affect individuals or society.
- A new governance authority would need capacity to establish requirements for identifying high-consequence AI and to conduct or support case-by-case reviews of submitted AI tools.
- Existing sector-specific regulators are described as capable of developing domain-specific frameworks, while a new authority could support foundational consistency across departments, agencies, and commissions.
- A standalone agency could create a federal locus of AI governance expertise and institutional memory, but it could also add costs, duplicate sector-specific reviews, and face limits coordinating across independent and executive agencies.
- Locating a new office within an existing department or agency could reduce setup burdens and use existing authorities, but legacy missions might shape the office and weaken perceptions of neutrality across government stakeholders.
This options memo argues that AI should be governed by sector and outcomes, with regulation focused on high-consequence use cases that may benefit or harm individuals and society. It identifies two regulatory functions: setting requirements or guidance to identify high-consequence AI use cases, and applying regulation to AI tools deemed high consequence. The memo states that existing sector-specific regulatory bodies can develop domain-specific frameworks, while a new governance authority could either regulate directly or support regulators by promoting foundational consistency and managing case-by-case review submissions. It compares three possible locations for that authority: a standalone independent agency, a new office in an existing department or agency, and a new office or unit within the Executive Office of the President. It then assesses four staffing models: full-time permanent staff, permanent staff supplemented by rotating fellows, permanent staff supplemented by contract experts, and permanent staff supported by a federally funded research and development center. Across the options, the memo highlights tradeoffs involving expertise, institutional memory, setup and operating costs, conflicts of interest, regulatory capture concerns, interagency authority, and the risk of cumbersome or duplicative review processes. Its policy relevance lies in structuring choices for federal AI governance capacity without prescribing a single institutional model.
Artificial intelligence should be governed by sector and outcomes.
SCSP
Published by the Special Competitive Studies Project. Co-publishers and external authors, where they exist, are credited in the byline above.