ANALYSIS Washington is using its G20 host role to lock in a deregulatory default for AI governance worldwide, even as the EU tightens its own rules and an OpenAI executive publicly calls for the one form of cooperation the framework conspicuously omits: bilateral safety talks with Beijing.
Why it matters
The "Carolina Principles" introduced at this week's G20 innovation ministerial in Chapel Hill, North Carolina, are not a treaty or a binding accord. They are a normative bid: an attempt by the United States to set the baseline assumption that AI should be governed through existing regulatory structures, not new ones5,6. If adopted broadly, the framework would constrain the policy space available to governments considering EU-style, technology-specific oversight. That makes the Chapel Hill meeting less a technical workshop than a standard-setting exercise with commercial consequences for every major AI lab.
The big picture
U.S. tech adviser Michael Kratsios, co-hosting the two-day meeting, told G20 officials that "policymakers do not need to approach each innovation in isolation and should not treat every emerging technology as a first-of-its-kind policy problem". Countries signing the principles agreed to "reserve new regulation for novel considerations" and to invest in "foundational research to accelerate discovery"14. Kratsios told reporters that the Chinese government signed the principles8. The Chinese embassy did not immediately respond to a request for comment.
The gathering drew a roster calibrated to reinforce the message. Google DeepMind's Demis Hassabis, Meta's Mark Zuckerberg, and SpaceX CEO Elon Musk addressed the ministers by video on Tuesday7. OpenAI CEO Sam Altman and Nvidia CEO Jensen Huang are scheduled to appear separately before delegates with Commerce Secretary Howard Lutnick on Wednesday12.
ANALYSIS The speaker list itself functions as a policy argument: the executives whose companies stand to gain most from lighter regulation are the ones briefing the ministers who would write it.
Reuters noted that the U.S. goals "largely align with the thinking of the world's biggest AI companies, nearly all of which are American," and that federal requirements "could hurt the industry's profits if they slow the release of new models or prompt the companies to change how their products perform to address security concerns".
Meanwhile, the EU continues to pull in the opposite direction. Al Jazeera reported that the European Union and the United States "continue to pull in opposite directions on artificial intelligence, with Washington urging other governments to loosen constraints"3. ANALYSIS The Carolina Principles are, in effect, a counter-programming effort against the EU AI Act's model of prescriptive, risk-tiered regulation.
Between the lines
The most striking tension sits not between Washington and Brussels but within the U.S. policy ecosystem itself. Days before the Chapel Hill meeting, Dean Ball, OpenAI's head of strategic futures, wrote on X that he has "always believed US/China AI safety collaboration would be desirable but thought it was unlikely to happen"1. His call came ahead of a planned summit between President Trump and Chinese leader Xi Jinping, and arrived as Chinese state media published what the South China Morning Post described as a "scathing attack on American frontier governance in artificial intelligence".
ANALYSIS The Carolina Principles framework says nothing about bilateral safety cooperation with Beijing. China's signature on the principles, as reported by Kratsios, addresses the commercial regulation question but leaves the safety-cooperation question entirely open. Ball's public appeal from inside OpenAI suggests that at least some within the leading U.S. lab see a gap between the administration's trade-oriented governance posture and the safety architecture they consider necessary.
A United Nations panel recently warned that AI developments are outpacing scientific understanding and government policy13. Bill Gates published an essay last week arguing that "even under the best circumstances, the transition to this new AI era will be one of the most turbulent times in human history"4. ◆ Both statements sit uncomfortably alongside a framework whose central instruction to governments is restraint.
Chinese open-weight AI models are becoming increasingly capable compared with proprietary systems from American companies such as Anthropic and OpenAI. ◆ That competitive dynamic gives Washington a dual incentive: keep regulation light enough for U.S. labs to move fast, while avoiding a governance vacuum that lets capable open-weight models circulate without oversight.
What's next
The Chapel Hill meeting is one of a series of gatherings taking place over the next few months in U.S. cities leading up to the G20 leaders' summit in Miami in December. Wednesday's session, featuring Altman, Huang, and Lutnick, will test whether the Carolina Principles attract additional signatories or remain a statement of U.S. preference. The planned Trump-Xi summit offers a separate track where the safety-cooperation question Ball raised could surface at the head-of-state level. For now, the U.S. has placed its bet: governance by restraint, sold to the world's largest economies by the executives who benefit most from it.