G-20 Adopts US-Backed Carolina Principles Limiting New AI Regulation

The G-20 adopted the Carolina Principles, a framework the U.S. delegation brought to the September 2 innovation ministerial in Chapel Hill, North Carolina. Commerce Secretary Howard Lutnick said all 20 members, including China, agreed. The framework urges governments to reserve new AI regulation for “novel considerations,” while the European Commission sent information requests to more than 30 AI companies during the meeting.

The Carolina Principles

The consensus statement says emerging technologies have the potential to increase productivity and expand economic opportunity. Its six pillars cover pro-innovation policy, technology for opportunity and prosperity, technical workforce development, intellectual property for AI, standards, and industrial investment in supply chains.

The Carolina Principles ask countries to invest in research, strengthen paths from invention to commercial use, and support trusted technology adoption. The ministerial also produced AI Prosperity Objectives and an AI Prosperity Compact aimed at workforce development and private-sector partnerships.

Lutnick called agreement by all 20 countries “an enormous amount of work.” China signed with the other members; Lutnick and Michael Kratsios, director of the White House Office of Science and Technology Policy, met Chinese officials.

The language “reserve new regulation for novel considerations” comes from a description of the draft framework; the announcement does not say what counts as novel.

Washington’s regulatory case

Kratsios said policymakers should not treat each emerging technology as a first-of-its-kind policy problem. The Trump administration has paired that position with voluntary prerelease testing, government reviews of the most powerful models, and a brief June 2026 shutdown of two Anthropic models over security concerns.

Nvidia Chief Executive Jensen Huang spoke in person at a fireside chat moderated by Lutnick, who introduced him as “an American treasure.” Advances in model training were already making systems safer than rules proposed by the Biden administration and others, he said.

“Don’t regulate hypothetical, theoretical harm. Regulate actual and pragmatic harm,” Huang told the ministers.

OpenAI Chief Executive Sam Altman also presented adoption as unavoidable. “It is non-negotiable. You have to use it,” he said. Yet Altman differed from Huang on risk, telling ministers, “Worry is warranted. Worry is how we anticipate problems.” He said global standards and oversight still had a place, while self-regulation could help companies avoid catastrophic risks.

Brussels begins compliance checks

On Tuesday, September 1, as the Chapel Hill meeting opened, the European Commission confirmed that it had sent information requests to more than 30 AI companies worldwide. The requests focus mainly on safety and copyright compliance. They are preliminary inquiries, not formal investigations, but can lead to enforcement proceedings under the AI Act.

Henna Virkkunen, the Commission’s Finnish executive vice-president for tech sovereignty, attended the ministerial and rejected the claim that regulation and innovation are opposites. She said European law addresses conduct in advance, while U.S. courts often examine the same companies and practices after disputes arise.

Eighty percent of European respondents in a recent Eurobarometer survey said careful AI regulation was important for safety, Virkkunen told AFP in a sideline interview Wednesday. “For Europeans, it’s important that we can trust these technologies,” she said.

She also pointed to hundreds of U.S. state-level AI rules. The EU’s approach, she said, was designed to give companies one set of rules across the bloc. Enforcement powers under the AI Act took effect in August 2026, after most leading developers had cooperated voluntarily.

Governments seek a middle course

Chris McDonald, Britain’s science, innovation and investment minister, said company leaders and national governments carry different responsibilities. Britain wants faster AI adoption without giving up intellectual-property protection or public trust.

“The porridge can’t be too hot or too cold. It’s got to be just right, and that’s a matter of judgment,” McDonald said.

Dozens of students and local activists protested outside the meeting on Wednesday. A separate estimate put the crowd at 100 to 200 people. The host county had imposed a one-year moratorium on new data centers amid concern about electricity and water use. Calling the moratoria misinformation, Lutnick singled out misleading water-use estimates and said “the next set of opportunities just won’t go to you.”

The infrastructure demand

Huang told ministers that countries should treat AI capacity like roads, electricity, and the internet. “Every single country needs to build infrastructure so that you can support your own local economy,” he said.

Meta Chief Executive Mark Zuckerberg told the Tuesday session that the data-center buildout would require hundreds of thousands, and perhaps millions, of skilled-trades jobs. Meta could not find enough workers to build the facilities it wanted, he said.

Elon Musk predicted on Tuesday that there would be a significant power shortfall next year, “not (the) distant future.” He said his company was building generation capacity alongside its data centers. Altman said national governments could rent computing capacity abroad instead of building facilities.

The next test

The United States holds the rotating G-20 presidency in 2026, and the ministerial meetings are part of the run-up to a leaders’ summit that Trump will convene on December 14 and 15 at his Trump National Doral resort in Miami. Before then, the United States and China are expected to discuss AI cooperation in talks later this month involving Trump and Chinese leader Xi Jinping.

Anthropic Moves Claude Monitoring Data Into Customer Clouds After Enterprise Pushback

Anthropic is moving enterprise misuse-monitoring data into customers’ own cloud environments after the 30-day retention rule introduced with Fable 5 in June 2026 made the model difficult for some regu

The Implicator

Abbott and Shapiro Restrict Data Centers After Courting $60 Billion

In November, Texas Gov. Greg Abbott sat beside Google chief executive Sundar Pichai at a data center in Midlothian as Google announced plans to invest $40 billion in data centers across the state. Abb

The Implicator

Trump Calls Rejecting Data Centers a ‘Mistake’ Despite 61% Opposition

President Donald Trump said communities that reject data centers are “making a mistake” in an interview that aired in full on Sunday, even as 61% of U.S. adults opposed a new data center in their area

The Implicator

Marcus Schuler

San Francisco

Editor-in-Chief and founder of Implicator.ai. Former ARD correspondent and senior broadcast journalist with 10+ years covering tech. Writes daily briefings on policy and market developments. Based in San Francisco. E-mail: editor@implicator.ai