G20 ministers endorsed the nonbinding “Carolina Principles” that call for using existing laws before creating AI-specific requirements.
The United States used a G20 innovation meeting it hosted to promote a hands-off approach to regulating AI and other emerging technologies.
G20 ministers agreed that governments should rely on existing sector laws where possible and create new regulations only for problems those laws cannot adequately address.
The position is part of the Carolina Principles for Emerging Technologies, which ministers adopted by consensus at a two-day meeting in Chapel Hill, North Carolina.
The principles are not binding and do not create new requirements for companies. They leave each G20 member in control of its own technology policies.
G20 leaves AI rules to national governments
The G20 ministerial statement supports regulating AI according to the risks created by its use in a particular sector. However, it does not establish common G20 requirements for AI developers or companies using AI. It creates no shared licensing system, safety-testing rules, or enforcement body.
Instead, the statement says consumers should determine through free-market competition which AI models best meet their needs. It calls on governments to promote access to secure and trustworthy AI models and says standards should not create unnecessary barriers to international trade.
The agreement also leaves each G20 member responsible for its own AI policies. This allows the United States and the European Union to endorse the same statement while maintaining different regulatory systems.
The Carolina Principles provide the broader regulatory approach behind the agreement, but they apply more broadly to emerging technologies.
Those principles tell governments to rely on existing sector laws where appropriate and create new regulations only when those laws leave gaps. They also support supervised testing programs that may offer exemptions from selected regulations and faster approvals for pilot projects.
The documents do not identify which existing laws should apply to AI, which AI regulations could be waived during testing, or when an AI-specific rule would be necessary.
Michael Kratsios, director of the White House Office of Science and Technology Policy, described the agreement as supporting “flexible policy frameworks crafted to promote innovation.”
EU has already adopted binding AI rules
The U.S. position contrasts with the EU AI Act, which already imposes binding AI rules across the bloc, including in G20 members France, Germany, and Italy.
The EU AI Act became broadly applicable on August 2. It bans specified AI uses and places transparency, copyright, safety, and security obligations on providers of general-purpose AI models.
EU regulators can require companies to provide technical records, correct violations, or restrict AI systems. They can also impose fines for noncompliance. Some requirements for AI used in sensitive areas and regulated products will take effect later.
The European Union nevertheless joined the consensus. The statement does not require the bloc to weaken or replace its AI Act. It allows each G20 member to maintain its own policies and create technology-specific regulations when existing laws leave gaps.
The agreement also sets no deadlines for governments to adopt the Carolina Principles. Any changes to national AI rules will depend on decisions by individual G20 members.
