US Presses G20 to Adopt 'Carolina Principles': The Light-Touch AI Framework Musk, Altman, and Huang All Want

The most consequential AI policy fight of the year is not happening in Brussels or Washington. It is happening in Chapel Hill, North Carolina, where the United States is hosting a G20 innovation ministerial this week — and where the Trump administration is pressing the world’s largest economies to sign on to something called the Carolina Principles, a light-touch framework for governing artificial intelligence that would commit governments to avoiding the creation of new AI regulators altogether.

The stakes are straightforward: whether the world’s dominant AI regulatory model over the next decade looks more like the EU’s AI Act — ex-ante rules, prohibited uses, heavy documentation duties — or more like the US approach of voluntary commitments, sector-specific enforcement, and industry self-regulation. What makes the meeting unusual is who is in the room. Commerce Secretary Howard Lutnick is hosting fireside chats with OpenAI CEO Sam Altman and Nvidia CEO Jensen Huang, and Elon Musk appeared virtually alongside former White House AI czar David Sacks — a lineup that reads less like a diplomatic summit and more like a joint appearance by the technology’s biggest builders and its most powerful customer.

What the Carolina Principles actually say

According to advance reporting from Reuters and AFP, the US-backed framework centers on a few core commitments. Governments that sign on would pledge not to create new regulatory bodies dedicated to AI. Instead, oversight would remain with existing agencies enforcing existing laws — antitrust, consumer protection, product safety — applied to AI as they would be to any other technology. The principles also favor “lighter, collaborative regulation,” closer public-private cooperation, and interoperability across jurisdictions, with discussions extending to adjacent questions like crypto-asset trading rules.

It is, in effect, the anti-AI-Act. The EU’s regime classifies AI systems by risk tier and imposes obligations before products reach the market. The Carolina Principles ask governments to trust that existing legal machinery, plus industry self-policing, is enough — even as frontier models gain the ability to write code, execute multi-step tasks, and operate inside critical systems.

Musk: build data centers everywhere, and don’t regulate like Europe

Elon Musk’s virtual appearance delivered the bluntest version of the industry’s argument. Musk called on countries to build AI data centers of their own, according to The Information — a pitch that conveniently aligns with his companies’ ambitions in compute infrastructure. He urged the G20 to “tread softly” on regulating tech and AI, arguing that European-style rules inhibit innovation, and he defended the data-center buildout against mounting environmental and community opposition.

The irony was not lost on observers. xAI’s Colossus facility in Memphis has drawn lawsuits over emissions from unpermitted gas turbines — regulators, it turns out, only feel like innovation killers until your own data center needs a permit. Musk’s argument that countries should race to build compute while simultaneously asking governments to keep their hands off reflects the industry’s central bet: that the geopolitical competition for AI leadership is so intense that no government will risk slowing its own champions.

The counter-pressure: financial stability has entered the chat

The Chapel Hill summit opens at a moment when other parts of the regulatory system are sounding alarms in the opposite direction. Just a day before the ministerial, Bank of England Governor Andrew Bailey — chair of the Financial Stability Board — sent a formal letter to G20 finance ministers warning that frontier AI now threatens global financial stability, with AI-driven cyber risk as “the most immediate concern.” The FSB called for “appropriate steps to support safe and responsible model release and deployment on a global basis” — language that points toward exactly the kind of pre-deployment scrutiny the Carolina Principles are designed to avoid.

The two documents describe the same technology in mutually incompatible terms. One asks the world’s governments to promise never to build an AI regulator; the other asks whether frontier models should face release evaluations akin to bank stress tests. That tension will define the next year of AI governance: the FSB letter represents institutions already inside the system’s risk perimeter arguing for guardrails, while the Carolina Principles represent the builders arguing that any guardrail built too early becomes a permanent tax on innovation.

Why the venue matters

Hosting the ministerial in North Carolina is itself a message. The state has aggressively courted AI infrastructure investment, and Altman is expected to highlight OpenAI’s work in the region — a reminder that in the US model, AI policy and industrial policy are fused. When Lutnick hosts Huang, the subtext is chips; when he hosts Altman, it is compute; when Musk appears beside David Sacks, it is the ideological case for speed. The Carolina Principles would export that fusion as a template: grow the AI base first, govern it later, with whatever agencies already exist.

Whether the principles survive contact with actual G20 diplomacy is uncertain. The EU is unlikely to endorse language implying its AI Act was a mistake, and countries that built digital regulators in the last decade — and staffed them — will resist committing to never build another. But even a partial endorsement would give the light-touch camp a multilateral banner to rally under, and a counterweight to the Brussels model in third markets.

For the AI industry, the calculation is simple. The cost of the EU AI Act’s compliance regime is real; the cost of a fragmented world — where every jurisdiction invents its own AI authority — is worse. A G20-endorsed “no new regulators” principle would be the cheapest regulatory insurance money can buy. Whether it is also good public policy is precisely the question Chapel Hill is supposed to answer.