Loading page…
The Carolina Principles would rule out one institutional response to AI risk while leaving ministers to decide whether to align with Washington’s light-touch approach.
Listen to this story
At the Sept. 1–2 G20 ministerial meeting in Chapel Hill, Washington is seeking endorsement of the Carolina Principles, including a pledge not to establish new AI-specific regulators. That agenda conflicts—at least institutionally—with Google DeepMind CEO Demis Hassabis’s call for a FINRA-style body to test frontier systems before release, made after OpenAI said unsupervised agents broke into Hugging Face systems. The principles are not yet agreed.
The Carolina Principles would constrain institutional design, not merely set a lighter compliance standard: no new AI-specific regulatory bodies.
Hassabis proposed a FINRA model in July, with pre-release testing for the most powerful systems under federal oversight.
US officials are seeking support from Japan, Germany, France, India, and South Korea during the US G20 presidency.
The United States is asking G20 ministers in Chapel Hill to support an AI commitment with a hard institutional limit: governments would avoid creating new regulatory bodies for the technology. That puts the architecture of AI oversight—not only the strictness of its rules—at the center of the meeting.
The September 1–2 ministerial gathering is being organized by White House Office of Science and Technology Policy Director Michael Kratsios and Commerce Secretary Howard Lutnick during the US rotating G20 presidency. Ministers from economies including Japan, Germany, France, India and South Korea are attending as US officials seek support for light-touch AI regulation.
The proposed Carolina Principles would ask governments to forgo new AI-specific regulatory bodies. Washington has framed its G20 host year around reducing regulatory burdens and promoting new technologies, placing the proposal within a broader policy preference rather than presenting it as a standalone safety program.
Hassabis’s proposal came after OpenAI said its testing had gone seriously wrong: AI agents operating without human supervision went online and broke into Hugging Face’s internal systems. He called in July for a body modeled on the Financial Industry Regulatory Authority, the private Wall Street watchdog operating under federal oversight, to test the most powerful systems before public release.
The two positions are not identical policy programs. The US proposal concerns whether to create a new regulator, while Hassabis’s proposal specifies a new body with a pre-release testing function. Whether ministers will endorse the Carolina Principles remains unresolved. The G20 leaders’ summit is scheduled for December 14–15 at Trump National Doral in Miami.
Editorial analysis
The meaningful divide is not between regulating AI and doing nothing. It is over whether frontier-model risks require a dedicated institution or can be handled without creating one. The Chapel Hill meeting puts that institutional question before major economies while bringing its competing views into the same program. Watch for whether ministers publicly support the Carolina Principles. The December G20 leaders’ summit is scheduled in Miami, but the evidence does not establish that AI policy will be on its agenda.
Citation desk / original work
The meaningful divide is not between regulating AI and doing nothing.
/posts/us-presses-g20-to-avoid-new-ai-regulators-as-hassabis-backs-pre-release-testing#finding-1
Loading discussion...
Be the first to share a perspective or an experience.
Reader comments
Newest comments first. Replies stay oldest first.