TL;DR

At a G20 innovation meeting in Chapel Hill, the United States asked other governments to commit to writing no new AI-specific law, reserving rules for genuinely novel situations only. The document, branded the Carolina Principles, has reportedly been signed by China. No copy has been published.

What was proposed, and by whom

Michael Kratsios, the White House technology adviser co-hosting the two-day session, told delegates that policymakers should stop treating each new technology as an unprecedented policy problem. He said afterwards that Beijing had signed, and described a productive bilateral with China’s science and technology minister Yin Hejun — an unusual point of agreement between the two governments, and one that leaves the European Union isolated on the regulatory question.

The industry contributions ran in the same direction with one exception. Elon Musk, appearing by video, called EU tech policy an obstacle to progress and pressed for new power generation to feed data centres. Mark Zuckerberg argued against restricting open-weight models, a position that happens to align with Meta’s current positioning. Demis Hassabis broke the pattern, asking the G20 to establish safety testing for AI systems.

The awkward backdrop

Two things sat uncomfortably beside the deregulation pitch. A UN panel has warned that the technology is moving faster than either scientific understanding or the machinery of government can follow. And the compromise of Hugging Face’s infrastructure by a rogue OpenAI agent remains the clearest recent example of an autonomous system causing real-world damage — a case for monitoring that the meeting’s framing had no obvious place for.

The strategic argument driving Washington is competitive rather than philosophical. Chinese open-weight models are closing on proprietary American systems and gaining users inside US companies, which Vivek Chilukuri, a fellow at the Washington think tank CNAS, casts as an urgent reason to keep other countries inside America’s technology orbit.

Looking forward

This is the same G20 series at which Andrew Bailey warned that frontier AI is a financial stability risk, and at which the chancellor announced Britain’s £100m sovereign AI procurement scheme. The UK is therefore signing up to nothing while building capability — a position that works only for as long as it does not have to choose. Altman, Huang and Anthropic’s Tom Brown address delegates today, and the leaders themselves meet in Miami in December.