For the better part of three years, the AI policy world has been trying to hold at bay the idea that AI regulation would inevitably adopt the European Union’s regulatory model: binding, technology-specific, risk-calibrated, and ultimately global in reach. That assumption now has a rival. The G20 Innovation Ministerial, which concluded last week in Chapel Hill, North Carolina, advanced a free-market vision of guidance that AI CEOs hope becomes the new norm for AI governance.
The White House Office of Science and Technology Policy and the Commerce Department jointly hosted the event under Secretary Howard Lutnick, with Treasury Secretary Scott Bessent also participating. The two-day gathering concluded on September 2. In a unanimous vote that included China, all 20 members signed the Carolina Principles, as U.S. technology adviser Michael Kratsios dubbed them.
The framework is intentionally limited in scope. It holds that existing statutes should govern AI wherever they already apply; that new rules should be reserved for genuinely unprecedented problems; that oversight should remain dispersed across sector-specific agencies rather than consolidated under a single omnibus AI regulator; and that moving research into real-world deployment depends on public-private collaboration and blended financing. Kratsios articulated the underlying logic directly, arguing 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.”
The Carolina Principles carry no binding force, and the next formal checkpoint is the G20 Leaders’ Summit in Doral this December. China already operates one of the most restrictive AI environments anywhere, through its algorithm-registration and generative-content requirements. Endorsing a voluntary pledge that changes nothing about how China governs AI domestically costs nothing. It is a diplomatic win obtained for free, in a setting where refusing to sign would have made Beijing the odd one out. So they signed it.
The counterpoint to this aspirational document is that on August 2, the European Union’s AI Act brought its obligations for high-risk systems into force, and those obligations rest on something corporations understand better than principles: a risk-tiered legal framework with enforceable financial penalties. Companies found in violation of the Act’s prohibited-practices category can face fines of up to €35 million or 7 percent of global annual turnover, whichever is greater. So while one side is articulating a much lighter regulatory approach, the other just published a punishing financial schedule.
Unfortunately, this is where the merits of “light-touch” governance (which I support) meet the reality that capital investors want to know the risk before investing.
Physical compute and power, the skilled workforce, and investment dollars can shift to whichever jurisdiction imposes the least friction, and they will. Being right about a rule’s substance does not prevent a data center, a model-training operation, or a funding round from landing somewhere cheaper to operate. Regulators can win the argument and still lose the industry.
Setting diplomacy aside, the Carolina Principles point to a concrete stake. They call for applying laws already on the books, limiting new rulemaking to genuinely novel harms, and keeping oversight spread across sectors rather than housed in a single AI agency.
That is a modest, realistic request, and I think it is the right one. Washington does not need fresh AI legislation to make model developers answer for fraud, discrimination, or safety lapses that existing consumer-protection, sectoral, and liability law already cover. What the Principles lack is any enforcement mechanism to make that commitment durable, domestically or internationally, that will satisfy the current hunger for regulation by the states, Congress, and the EU.
That is precisely why the meaningful test is not the signatures collected this week but whether the G20 gathering in December produces something with real force and consequences. There is room for a joint process to review cases where current law falls short, rather than yet another restatement of good intentions.
The potential rewards of getting this right are concrete. Productivity gains from AI across medicine, energy, manufacturing, and defense could add meaningful percentage points to GDP growth over the coming decade, and the biggest beneficiaries will be the jurisdictions that attract frontier development rather than merely regulating it after the fact. Chapel Hill produced a sector-by-sector philosophy for regulating current technology for the G20. Brussels produced a detailed framework for financial fines. The question is which of those plans investors will view as the real regulator of where AI is heading.