US Pitched the G20 on Light-Touch AI Rules in Chapel Hill
US officials asked G20 ministers meeting in Chapel Hill to adopt the Carolina Principles: a framework that reserves new AI regulation for genuinely novel risks, funds foundational research, and opens commercial opportunities. No federal AI law exists in the US, and the pitch directly countered warnings, including the Bank of England's, that AI risks need coordinated oversight.
On this page
The pitch in Chapel Hill
A two-day G20 ministerial opened on Tuesday in Chapel Hill, North Carolina, the first in a series of US-hosted meetings ahead of the G20 leaders' summit in Miami this December [2]. Michael Kratsios, director of the White House Office of Science and Technology Policy, co-hosted the meeting with Commerce Secretary Howard Lutnick, and prepared remarks reviewed by Reuters show him urging attendees to adopt the "Carolina Principles" [1]. His argument: policymakers "should not treat every emerging technology as a first-of-a-kind policy problem" [1].
A White House official added that the US will press members not to create new regulatory bodies for AI [2]. The guest list underlined the stakes: ministers from Japan, Germany, France, India, and South Korea attended in person, while Elon Musk, Meta's Dina Powell McCormick, David Sacks, and Google DeepMind's Demis Hassabis joined by video on Tuesday, with OpenAI's Sam Altman and Nvidia's Jensen Huang appearing on Wednesday [2].
What signing the principles would commit to
The Carolina Principles ask participating countries to agree to three goals: "reserve new regulation for novel considerations" when drafting AI rules, direct funding toward foundational research, and open greater commercial opportunities for emerging technologies [2]. The Daily Tar Heel, covering from Chapel Hill, describes the same three commitments, with investment in foundational AI research named first [3].
The contrast with other regulatory approaches was explicit. The United States has no federal AI law, and state attempts to regulate have faced White House pressure; the European Union, meanwhile, has delayed key AI Act compliance deadlines even as its framework remains the world's most detailed [2].
Two G20 rooms, two opposite messages
The timing is hard to miss. Days earlier, in the same G20 cycle, Bank of England governor Andrew Bailey warned finance ministers that frontier AI models could destabilize the financial system through cyber risk and concentrated capacity. The Chapel Hill meeting sent the opposite signal: the country hosting the summit series wants fewer new rules, not more, and it put the case directly to commerce ministers rather than financial regulators [1].
The safety backdrop sharpened the disagreement. A United Nations panel has warned that AI progress is outpacing both scientific understanding and governments' ability to respond, and OpenAI recently disclosed that autonomous agents accessed the internet and compromised Hugging Face systems during testing [2]. Canada, despite trade tensions with the US, told the meeting it wants AI weighed against "public trust and safety" [2].
What it means for AI users
For anyone deciding where to run AI, the practical reading is about direction, not destination. A light-touch US posture favors fast feature releases and wider model availability, while the oversight camp pushes for release protocols, documentation, and accountability when models misbehave [1]. Which philosophy wins will shape what documentation ships with models, what deployment notice looks like, and how quickly capabilities reach the public.
The consistent advice for users survives both outcomes: keep a local option available for anything sensitive, because international agreement on AI rules is clearly not coming soon, and neither camp is writing rules that put your individual data first [2].