On Tuesday, Sept. 1, at the Carolina Inn in Chapel Hill, North Carolina, the United States asked the world's largest economies to agree that artificial intelligence does not need a rulebook of its own. The same day, the European Commission sent information requests to more than 30 AI companies asking whether they are complying with the rulebook it already has.
That is the transatlantic AI relationship in a single afternoon.
The two-day G20 Innovation Ministerial, hosted by the Commerce Department and the White House Office of Science and Technology Policy, closed Wednesday, Sept. 2, with a consensus statement and a framework the administration calls the Carolina Principles for Emerging Technologies. Its central claim is that governments should apply existing sector-specific law to AI and reserve new regulation for genuinely novel problems.
"Policymakers do not need to approach each innovation in isolation and should not treat every emerging technology as a first-of-its-kind policy problem," said Michael Kratsios, director of OSTP and a technology adviser to President Donald Trump.
The framework commits signatories to three goals: investing in foundational research, strengthening development and commercialization pathways, and enabling trusted adoption and deployment. It is non-binding, and it discourages standing up dedicated AI regulators. Nineteen countries plus the European Union and the African Union sent representatives; the roster included China and the Russian Federation. Commerce Secretary Howard Lutnick said at the close that the framework had been endorsed.
"Achieving consensus in the G20 is no small feat, but the topic of innovation as a driver of growth brought us together for a historic moment of unity," Lutnick said in the White House statement.
Brussels was busy the same week
The Commission's information requests, a preliminary step that can precede formal investigations, focus mainly on safety and copyright compliance, spokesman Thomas Regnier said. They were announced over the weekend by Henna Virkkunen, the Commission's executive vice president for tech sovereignty, in a LinkedIn post.
"Our goal is to ensure that AI in Europe is developed, released and used safely and transparently," Virkkunen wrote, adding that Brussels is "ready to take all necessary steps." She attended the Chapel Hill ministerial in person.
Europe's position is less absolute than the framing suggests. The AI Act's transparency obligations took effect Aug. 2, 2026. But five days earlier, on July 27, the Digital Omnibus on AI, Regulation (EU) 2026/1744, entered into force and pushed the AI Act's own high-risk deadlines back: standalone systems in areas such as employment, education and access to essential services now face Dec. 2, 2027, rather than Aug. 2, 2026, and AI embedded in regulated products gets until Aug. 2, 2028. High-risk violations still carry penalties of up to 15 million euros or 3 percent of global annual turnover.
"With simpler and innovation-friendly rules, we make it easier to innovate without lowering the bar on safety," Virkkunen said of that package.
The strongest version of each case
The deregulatory argument was made most bluntly by Elon Musk, who appeared at the ministerial alongside Meta's Mark Zuckerberg. Innovation, Musk argued, requires that entrepreneurs be relatively free of regulation, "meaning that new things must be default legal as opposed to default illegal." Of Europe he said: "Things are generally default illegal. It slows it down quite considerably." Zuckerberg said the data center build-out would require "hundreds of thousands, and maybe millions" of skilled tradespeople his company cannot currently find.
The precautionary argument rests on what has already happened. Al Jazeera reported that OpenAI acknowledged in July that its models had autonomously hacked into a coding platform during security testing, and that Anthropic disclosed the same month that its systems had gained unauthorized access to outside organizations during testing. Brussels' position is that transparency and testing obligations are the price of deployment at scale, not a tax on it.
Nor is industry united behind Chapel Hill. Demis Hassabis, co-founder and chair of Google DeepMind, who joined the ministerial remotely, has called for an independent, industry-funded US body, modeled on FINRA, the private regulator that polices Wall Street brokerages, to test the most powerful models before public release. That is close to the kind of technology-specific institution the Carolina Principles counsel against.
Why This Matters
Fragmentation is not abstract for anyone shipping software into both blocs. A company deploying a resume-screening model in Frankfurt and Charlotte now runs two compliance regimes off one codebase. In the EU, an Annex III high-risk system means conformity assessment, technical documentation, event logging, human oversight and post-market monitoring, on a December 2027 clock. In the US, it means whatever the sector regulator already required, employment law and the EEOC for hiring, banking regulators for credit, with nothing AI-specific layered on top.
The usual outcome is not two products. It is one product built to the strictest applicable standard, with Brussels setting the floor. That is the dynamic that made GDPR a de facto global default, and it is expensive, and it is also, from the Commission's perspective, precisely the point.
The Carolina Principles change that arithmetic in one respect. With 20 economies nominally aligned on a sector-specific default, including China, the EU can no longer describe its approach as the emerging global consensus. It is describing an exception. Firms that assumed a single worldwide compliance build would suffice may now find geofencing cheaper: ship in the US, delay in Europe, engineer to the EU standard only for the EU. Several American developers already stagger their European launches for exactly this reason, and the incentive to do so just got stronger.
Three things to watch. Whether the Carolina Principles survive the trip from a ministerial communique to the G20 leaders' summit in December, where they are expected to feed into discussions. Whether the Commission's information requests to more than 30 companies escalate into formal proceedings, which would be the first real enforcement test of the AI Act against American developers. And whether the deferred December 2027 high-risk deadline holds, or whether the omnibus that moved it once moves it again.
“Policymakers do not need to approach each innovation in isolation and should not treat every emerging technology as a first-of-its-kind policy problem.”— Michael Kratsios, Director, White House Office of Science and Technology Policy