At G20, U.S. and China Sign Light-Touch Regulatory Principles as EU Probes More Than 30 AI Companies the Same Day

On September 1, 2026, White House science and technology adviser Michael Kratsios chaired the G20 Science and Innovation Ministers’ Meeting in Chapel Hill, North Carolina, where he presented participating countries with a non-binding AI governance framework called the “Carolina Principles.” That afternoon, he announced at a press conference that the Chinese government had signed the document. At the same time, the European Commission in Brussels sent requests for information to more than 30 AI companies worldwide, a preliminary step that could lead to formal compliance investigations.

What the Carolina Principles Say—and What They Do Not Say

According to Reuters, the “Carolina Principles” contain three core commitments: limiting new regulatory actions to “genuinely novel” technological use cases; avoiding the creation of new dedicated regulatory agencies in the name of AI; investing in basic research to accelerate technological discovery; and lowering barriers to commercialization while expanding market opportunities for new technologies.

The document’s real informational weight lies in its omissions: there are no mandatory pre-market testing requirements for models, no compute thresholds, no incident reporting obligations, and no enforcement mechanism. In essence, it is a statement about “what not to do,” rather than an action plan for “what to do.” Kratsios’s central argument—that “policymakers need not treat every innovation in isolation, nor should they view every emerging technology as an unprecedented policy challenge”—is intended to persuade governments to rely on existing regulatory frameworks rather than build a new apparatus specifically for AI.

This logic directly reflects the interests of leading U.S. AI companies. Industry representatives attending the meeting included Tesla CEO Elon Musk, who criticized EU regulations on the spot as “impeding progress”; Google DeepMind co-founder Demis Hassabis; and Meta CEO Mark Zuckerberg, who spoke by video. OpenAI CEO Sam Altman and Nvidia CEO Jensen Huang were scheduled to meet the commerce secretary the following day.

Why China Signed: Reverse-Engineering the Interest Logic

Kratsios told reporters that he had held a “very good” bilateral meeting with Chinese Minister of Science and Technology Yin Hejun, and then disclosed that China had signed the principles. The Chinese embassy in the United States did not immediately respond to a request for comment, and Beijing did not publicly explain its reasons for signing.

Domestically, China has a well-developed AI regulatory system, with specific rules covering algorithmic recommendations, deep synthesis, and generative AI. At the international level, China’s support for “light-touch regulatory principles” can be understood through three layers of interest logic:

• Avoiding the spillover effect of EU standards. The EU AI Act is explicitly positioned as the “world’s first comprehensive AI regulatory framework,” and its provisions are designed to apply to companies providing services to EU users. If the EU model were to become the G20 consensus, overseas expansion by ByteDance, Alibaba, Baidu, and others would face compliance costs far more complex than those at home.
• Instrumental alignment in strategic competition. One of the core narratives advanced by the United States at this meeting was that AI competition comes primarily from China, and that U.S. companies must not be held back by regulation. China’s expression of support for this competitive framing carries a certain reverse-maneuvering logic: while acknowledging the existence of competition, it uses an “aligned” posture to gain bargaining leverage in the making of international rules.
• A diplomatic window for yielding softness against hardness. U.S.-China technological decoupling is still ongoing. Signing a non-binding statement carries extremely low costs, but it can serve as a signaling currency in bilateral relations: it does not affect China’s domestic AI policy, while still showing Washington a degree of willingness to cooperate.

The above are interest-based inferences by outside observers, not positions publicly stated by China. The “Carolina Principles” are a non-binding document, and signing them has no legal consequences in itself.

The EU’s Simultaneous Response Was No Coincidence

According to Al Jazeera, the European Commission sent requests for information on September 2 to more than 30 AI companies worldwide, focusing on two areas: safety compliance and copyright. Commission spokesperson Thomas Regnier explained that this was a preparatory step that could lead to formal investigations. EU Executive Vice-President Henna Virkkunen said Brussels was “ready to take all necessary steps” to advance enforcement.

The transparency provisions of the EU AI Act formally took effect in August 2026, and the Commission then sent inquiries to more than 30 companies at the very moment when G20 calls for light-touch regulation were at their peak. This was not an accidental procedural overlap, but the EU’s policy countermeasure to the “Carolina moment”: demonstrating its position through concrete enforcement action rather than through a declaration.

This response also reflects the EU’s structural dilemma in the G20 setting. France and Germany are both G20 members, but as EU member states, they are bound by the AI Act and, as a matter of law, cannot sign a document calling for “no new AI regulatory agencies”—precisely because the EU AI Office is such an agency.

The Industry’s Real Position: The Arrival of a Dual-Track Compliance Era

For companies actually operating cross-border AI businesses, this split creates a practical problem: regulatory compatibility is disappearing. In recent years, companies could bet that “EU standards would eventually spread into the global baseline.” That assumption has now been explicitly rejected.

The U.S.-led light-touch regulatory framework and the EU AI Act point in opposite directions across several key dimensions. The former advocates handling AI applications under existing laws, while the latter establishes dedicated AI risk classifications and lists of obligations. The former explicitly opposes the creation of new regulatory agencies, while the latter has already established the EU AI Office and begun enforcement. The former encourages rapid commercialization, while the latter requires high-risk systems to complete compliance assessments before entering the market.

Google DeepMind co-founder Hassabis proposed at the meeting the creation of an independent AI testing body similar to the U.S. Financial Industry Regulatory Authority (FINRA)—a proposal that directly contradicts the core commitment in the “Carolina Principles” not to establish new regulatory agencies. This shows that even within the U.S.-led camp, there are substantive disagreements over the boundaries of light-touch regulation.

Independent Assessment

The greatest significance of the “Carolina Principles” lies not in what they stipulate, but in what they confirm: the United States has abandoned the intention of building a unified AI governance framework at the G20 level, and has instead chosen to actively fragment global rules. China’s signing of the document is a low-cost, highly flexible strategic gamble. It does not bind domestic policy or require any legal amendments, yet it positions China on the “looser” side of the international regulatory contest.

The EU’s response reveals a deeper structural contradiction: Brussels’s AI Act imposes real constraints, while the Carolina Principles are a non-binding statement. At the enforcement level, the EU holds the stronger cards—it can fine any AI company entering the EU market, while the United States currently has no comparable measures it can take against allies that refuse to sign non-binding principles.

What will truly determine the future global AI regulatory landscape is not how many names appear on any given statement, but the extent to which the EU’s inquiries to more than 30 companies ultimately turn into substantial fines, and how U.S. AI companies weigh their revenues in the EU market. When compliance costs become high enough, commercial pressure will push companies toward EU standards regardless of what Washington’s advocacy documents say. This mechanism has already played out once with the GDPR.