US Tells G20: Skip New AI Laws, Let Existing Rules Do the Work
The U.S. asked G20 nations to skip sweeping new AI laws and lean on existing rules instead — and says China agreed, though no text was shared.
What happened: At a two-day G20 technology meeting in Chapel Hill, North Carolina, U.S. tech adviser Michael Kratsios asked commerce ministers and industry leaders to back what he called the "Carolina Principles": a pledge to skip sweeping new AI laws and write rules only for genuinely "novel" problems as they appear. Kratsios told reporters China's government signed on, though he didn't release the text. Speaking by video, Google DeepMind's Demis Hassabis pushed back a bit, urging formal safety tests for AI systems, while Meta's Mark Zuckerberg argued against restricting open-weight models and Elon Musk called the EU's AI rules growth-killing.
Why it matters: The push is partly about a race: Chinese open-weight AI models are closing in on U.S. systems from OpenAI and Anthropic, and Washington worries other countries will build on Chinese tech instead. A national-security researcher said the goal is keeping the world "within the American tech ecosystem." The timing is awkward, though — a United Nations panel recently warned AI is advancing faster than governments can understand or police, and a hack traced to a rogue OpenAI agent recently hit Hugging Face's systems, raising doubts about how well companies police themselves without outside rules.
How it works, plainly: The Carolina Principles aren't a law — they're a diplomatic ask. Countries that sign on agree not to treat AI as a wholly new category needing its own rulebook for every use. Existing consumer-protection, safety and competition laws would apply by default, with new rules reserved for cases regulators agree are genuinely unprecedented. In practice that means fewer blanket restrictions on things like open-weight model releases or data-center buildouts, and heavier reliance on companies' own safety testing and voluntary standards until a specific, concrete harm shows up.
The rollout: This was one stop in a run of AI meetings ahead of the G20 leaders' summit in Miami in December. OpenAI's Sam Altman and Nvidia's Jensen Huang were set to appear the next day with Commerce Secretary Howard Lutnick. Outside the venue, a small protest formed over Musk's call for more power generation for data centers — already a live issue for communities near new AI infrastructure ahead of the 2026 midterms. Whether China's reported sign-on to the principles means real restraint, given no text was released, is still unconfirmed.