US pushes G20 ‘Carolina Principles’ to curb new AI regulation

The Carolina Principles are Washington’s clearest attempt yet to define a global default for AI governance that keeps regulation light and predictable. By asking G20 countries to avoid creating new AI‑specific regulators and instead limit fresh rules to genuinely novel risks, the US is aligning its diplomatic posture with the commercial interests of OpenAI, Anthropic, Google DeepMind, Meta and Nvidia, all of whom have a lot to lose from fragmented, heavy‑handed regimes. The reported fact that China signed the principles gives this more weight than a standard talking point, even if the commitments are nonbinding.

In the race to AGI, this matters less for what it changes overnight and more for the baseline it tries to lock in. If major economies accept that AI should be governed through minimal new institutions, frontier labs gain more freedom to iterate quickly and ship powerful models, provided they can frame risks as manageable within existing law. At the same time, you can already see the tension on stage: DeepMind’s Demis Hassabis calling for serious pre‑release safety tests, Zuckerberg pushing aggressively for open‑weight models, and Musk attacking EU‑style regulation while warning about power‑hungry data centres.

The Carolina Principles will not settle those debates, but they shift the Overton window toward permissive innovation. That could pull marginal timelines for AGI deployments forward, especially in countries that were on the fence about building strong, independent AI regulators.