G20’s Carolina Principles: What the Chapel Hill AI Policy Deal Means

The U.S.-hosted G20 Innovation Ministerial in Chapel Hill, North Carolina, ended on September 2, 2026 with a consensus statement that put an innovation-first approach to emerging technology at the center of international policy discussions. Its most closely watched element was the Carolina Principles for Emerging Technologies, promoted by White House science and technology adviser Michael Kratsios. 16

The practical result was not a global deregulatory treaty or a prohibition on AI rules. Rather, it was a non-binding framework urging governments to use regulation carefully, rely on existing frameworks where appropriate, and create new rules for genuinely novel issues. 26

What the Carolina Principles say

The official ministerial statement presents the Carolina Principles as a way to promote innovation while supporting secure and trustworthy technology. They call on members to:

• invest in foundational research, domestic technological capability, entrepreneurship and the removal of research-and-development bottlenecks;
• speed secure testing, validation, commercialization and adoption of emerging technologies; and
• favor proportionate, adaptable, risk-based policy approaches consistent with domestic law and national circumstances. 6

The key AI-governance message was restraint in creating new rules. Kratsios told ministers that signatories would “reserve new regulation for novel considerations,” rather than treat every use of AI as requiring an entirely separate regulatory regime. 23

That distinction matters. The statement recognizes that regulation can be appropriate, but does not prescribe a new international AI regulator or remove a country’s ability to regulate AI under its own laws. 6

Who supported the framework?

The strongest confirmed outcome is ministerial consensus around the final Chapel Hill statement, which incorporates the Carolina Principles. 16

Reporting also said that China backed the non-binding principles, with China’s science and technology minister Yin Hejun identified as supporting or signing on to the framework. That alignment was politically significant because the U.S. and China are competing for AI leadership. However, the material supplied does not provide an authoritative, complete list of individual national signatories or detailed public remarks from Yin. 24

Why large AI companies favored a lighter-touch approach

For major AI developers, a policy preference for existing sector-specific rules and narrowly targeted new regulation can mean less duplication, lower compliance burdens and fewer delays before products reach users. It also supports rapid investment in computing capacity, research and commercial deployment.

That is the appeal of the U.S. position: AI firms can argue that overly broad, technology-specific rules risk slowing useful applications before regulators have identified a concrete harm. The counterargument is that frontier systems may create risks that older sectoral frameworks do not fully address. The consensus statement itself reflects this tension by pairing pro-innovation language with references to secure, trustworthy and risk-based deployment. 6

What key officials and executives said

Michael Kratsios was the leading advocate for the Carolina Principles. His pitch was that governments should encourage research, commercialization and adoption while limiting new regulation to novel considerations. 23

Yin Hejun was reported to have supported the non-binding principles for China. The supplied sources do not substantiate detailed comments from him on safety testing, open-weight models or the wider regulatory debate. 4

Tom Brown, Anthropic’s co-founder, was among the technology leaders scheduled to address the gathering. The available evidence does not support assigning him a specific Chapel Hill position on AI regulation or open-weight models. 1423

Demis Hassabis, Google DeepMind’s chief executive, took a more safety-focused position than the broad hands-off message. Reuters reporting republished by Yahoo said he called on G20 officials to establish safety tests for AI systems. 23

Mark Zuckerberg, Meta’s chief executive, emphasized the need for more AI data centers and pointed to a shortage of skilled workers needed to build them. Reporting also said he argued against restrictions on open-weight AI models. 2123

Elon Musk, speaking for SpaceX and Tesla, argued that AI innovation should be relatively free of regulation. He criticized European regulation as a drag on innovation and warned that inadequate electricity supply for AI infrastructure could give China an advantage. 21

Safety testing remains the unresolved issue

The Chapel Hill outcome did not settle the central AI-policy dispute: whether governments should act before the most capable models are released, or primarily address harms through existing laws and after concrete risks emerge.

Hassabis’s call for safety tests illustrates why the Carolina Principles should not be read as a universal industry consensus. Supporters of a lighter-touch approach emphasize innovation, deployment and competitiveness. Safety advocates argue that powerful systems may require testing and oversight tailored to their capabilities, particularly where models can act with limited human supervision. 23

China, open-weight models and the competition for AI influence

The U.S. push also had a geopolitical dimension. Reuters reported that increasingly capable Chinese open-weight models were strengthening the alternatives available to countries and companies outside the U.S. technology ecosystem. 3

For Washington, keeping partners tied to American AI products, infrastructure and standards is therefore about more than regulatory philosophy. An innovation-led policy pitch can be part of a wider effort to make U.S. technology the more attractive ecosystem—while China’s reported support for the principles demonstrates that competition and policy alignment can coexist on selected issues. 23

Data centers made AI policy an infrastructure debate

AI policy at Chapel Hill was also inseparable from physical infrastructure. Zuckerberg and Musk urged ministers to support more data centers and the electricity needed to run them; Zuckerberg cited construction-workforce constraints, while Musk warned of a power shortage. 21

That emphasis comes as data-center expansion faces public scrutiny over grid capacity, local impacts and energy costs. The ministerial therefore linked three issues often discussed separately: the speed of AI deployment, the rules governing it, and the energy and construction capacity required to sustain it.

Bottom line

The Chapel Hill meeting gave the U.S. an important diplomatic win: G20 ministers endorsed a shared, innovation-first framework rather than a new global AI regulator. But the Carolina Principles are non-binding, preserve national policy discretion, and leave the fundamental argument over frontier-model safety testing unresolved. 16