Washington used the first day of the G20 Innovation Ministerial in Chapel Hill to ask governments to avoid new organizations dedicated to AI regulation [E1]. White House technology adviser Michael Kratsios offered the “Carolina Principles,” under which participating countries would avoid entirely new AI regulations and focus fresh rules on “novel” situations [E1]. Under the proposal, governments would send familiar risks to existing authorities and reserve fresh rules for problems specific to AI [E1][E2]. The American hosts scheduled the ministerial across September 1–2, leaving the meeting unfinished at the September 1 freeze [E3].
Commerce’s public account showed Secretary Howard Lutnick meeting Poland’s economy minister on the ministerial’s sidelines to discuss trade and AI cooperation [E4]. Kratsios later said China had signed the principles, and he supplied no copy of the document [E1]. A reported country endorsement is not a G20 communiqué [E1][E3]. The Carolina language therefore remained Washington’s offer at the freeze [E1][E3][E4].
Washington had already laid down the same institutional preference at home [E2]. A March White House framework told Congress to keep AI applications with existing sector regulators and explicitly rejected a new federal rulemaking body [E2]. By carrying that formula to Chapel Hill, American officials turned a domestic design choice into a foreign-policy offer [E1][E2]. They asked partners to accept the architecture before every future risk had a name [E1][E2].
Commerce and the White House science office invited ministers from more than 20 countries to a two-day meeting [E3]. They said the ministerial would “seek to identify policy principles” for AI and other emerging technologies [E3]. Participating governments were still considering those principles at the September 1 freeze [E1][E3]. A common ministerial text would settle whether partners accepted Washington’s language [E3].
Under the American formula, banking supervisors could handle AI in finance, and health authorities could handle clinical systems [E2]. A dedicated AI authority would remain off the organizational chart [E1][E2]. Companies could move from development to deployment more quickly because they would face institutions they already know [E1][E2]. Governments would also retain wide discretion over what counts as a genuinely “novel” situation, the hinge on which future intervention would turn [E1].
Large American AI companies stand to benefit from Washington’s design [E1]. Washington’s position matches their preference for fewer new rules as their products spread abroad [E1]. Governments adopting the same principle would limit the number of bespoke agencies those companies must answer to [E1][E2]. Washington is using the Carolina Principles to join regulatory diplomacy to the contest over who writes the operating terms for AI [E1][E2].
Any final ministerial statement may narrow or omit the proposal after the scheduled second day [E3]. G20 consensus depends on that common text despite any country-level endorsements [E1][E3]. Through the September 1 freeze, the public record showed Washington pressing the case and American officials working the room [E1][E4]. Washington’s cleanest regulatory victory would be an empty chair: the AI authority that never gets built [E1][E2].