G20 innovation ministers meeting in Chapel Hill, North Carolina agreed a framework on 1 September that reads less like a regulatory programme than an argument against one.
Three pillars
The Carolina Principles commit signatories to three things: reserve new regulation for genuinely novel situations rather than writing fresh rules for each AI capability; put public money into foundational research rather than compliance infrastructure; and remove barriers to deployment and commercialisation.
Michael Kratsios, the White House science and technology adviser, put the reasoning plainly: policymakers “do not need to approach each innovation in isolation and should not treat every emerging technology as a first-of-a-kind policy problem”.
What is not in the text
There is no enforcement mechanism, no implementation timeline, and no stated measure of success. The framework does not say whether it is binding — which, for a ministerial statement, usually answers itself.
Nor is there a full signatory list. China was confirmed as a signatory at the close of the meeting. The other participants were not publicly confirmed, which for a G20 output is unusual enough to be worth stating plainly rather than assuming.
Where it sits against Brussels
The principles run in the opposite direction to the EU AI Act, which has been enforceable since 2 August and applies escalating obligations by risk tier. Chapel Hill’s first pillar — regulate only what is genuinely novel — is a direct rejection of that architecture.
Both approaches now exist, both claim to serve innovation, and companies deploying across both jurisdictions will comply with the stricter one. That is usually how these divergences resolve in practice, whatever the communiqués say.
What to watch
The signatory list. A framework China has signed and the European Commission has not is a different object from a G20 consensus, and until the full list is published it is not possible to say which one this is.
