US officials at the G20 Innovation Ministerial in Chapel Hill are pushing the Carolina Principles to keep AI regulation voluntary, setting up a clash with the EU ahead of December's summit.
The Trump administration is using this week's G20 Innovation Ministerial in Chapel Hill, North Carolina, to sell the world on what officials are calling the Carolina Principles, a "pro-innovation framework" designed to keep artificial intelligence policy voluntary, industry-led and out of the slow lane. The pitch landed Monday with sharp framing: any rules that throttle the sector would be a mistake.
Michael Kratsios, director of the Office of Science and Technology Policy, used his opening remarks to argue that "slow policy adaptation makes it harder to capture the full benefits of innovation." He warned that regulations "built for a different technological era prove poor fits for rapidly evolving, generally applicable inventions," leaving "development efforts abandoned, growth unrealized and nations falling behind." The message tracks with the White House's broader deregulatory posture across tech.
A heavyweight lineup is on hand to reinforce it. David Sacks, the former AI czar and co-chair of the president's council on science and technology, told attendees it would be a "disaster" if the US adopted a version of the "FDA for AI." The phrase echoes a wider Republican argument that pre-market approval regimes would entrench incumbents and freeze out smaller developers. Reporting from CNET details the official statements and the framing battle inside the room.
The framework
Three pillars anchor the proposal. The first is research support: federal backing for foundation-model work, compute access and academic-industry partnerships. The second is commercialization pathways, meaning lighter-touch procurement rules, faster permitting for data centers and export financing for US AI vendors. The third is "industry-led standards" for safety testing, evaluation and disclosure, a deliberate rejection of binding rules in favor of voluntary commitments.
Ministers from India, China, Russia and the EU's largest economies are attending, and the EU's AI Act remains the obvious sticking point. Brussels has spent three years building a risk-tiered regime with enforcement powers; the Carolina Principles read as a counter-architecture. Any language that survives the week feeds into the G20 Leaders Summit in Miami this December, which is where the political cost will be paid or avoided.
What is actually new
The deregulatory thrust is not new. What is new is the venue and the timing. Hosting the forum in Chapel Hill, a university town within driving distance of the Research Triangle, sends a geographic signal: the administration wants AI policy framed as an industrial strategy question, not a consumer protection question. Kratsios's line about "nations falling behind" borrows directly from the semiconductor playbook, where CHIPS Act subsidies have been justified in identical language.
The urgency is also economic. OpenAI, which now serves more than 900 million weekly active users on ChatGPT, told employees in March that it is "orienting aggressively" toward high-productivity use cases ahead of a potential IPO as soon as the fourth quarter, according to CNBC. A binding global rule book would complicate that pitch to public-market investors. So would any US analogue that locked in compliance overhead before a single quarter of post-IPO earnings.
Meanwhile, the products keep shipping regardless of what ministers decide. Google this week rolled out Guided Vision inside Gemini Live, giving the assistant the ability to describe scenes through a phone camera and walk users toward objects, menus or fine print. Google also confirmed AI Mode has crossed one billion monthly users, per coverage in the Manila Times, and quietly opened Gemini inside Google Classroom to K-12 students of all ages, according to Education Week, a move that left districts roughly 48 hours to decide whether to block it.
Why it matters for builders
For developers and product teams, the practical stakes are not the communique text but what gets imported into procurement contracts and export licenses afterward. A framework that prioritizes "industry-led standards" tilts the field toward consortia such as the Frontier Model Forum and NIST's AI Risk Management Framework, both US-anchored. Companies based in jurisdictions that adopt binding rules will still have to comply at home; the question is whether they can ship the same model globally or fork it by region, the same workaround that GDPR forced on data products a decade ago.
OpenAI's own roadmap offers a useful parallel. A tracker maintained by Index Lab shows the parts of its 2025 plan that monetized attention shipped early while the parts that required owning transactions or hardware quietly slipped. Voluntary standards tend to reward whichever incumbent shows up to write them.
FAQ
Will the Carolina Principles become binding?
No. The framework is non-binding on its face and feeds into the December Leaders Summit as guidance. Individual jurisdictions remain free to legislate.
How does this differ from the EU AI Act?
The EU regime is risk-tiered and enforceable, with fines for non-compliance. The Carolina Principles push safety testing, evaluation and disclosure into voluntary industry commitments.
What happens at the December G20 Leaders Summit in Miami?
Ministers' agreements this week become draft language for heads of state. Expect a fight over whether "AI safety" or "AI opportunity" anchors the final communique, and whether the EU's language survives intact.
Does this affect US startups or only big tech?
Both, but differently. Voluntary standards tend to entrench players who can fund the working groups. Smaller developers benefit from lighter compliance overhead but lose leverage if a single industry-led benchmark becomes a de facto gatekeeper for enterprise procurement.
