On September 2, 2026, all twenty G20 nations unanimously endorsed a framework called the Carolina Principles, backed by the United States, calling for restraint on new AI-specific rules. At the same time, the EU AI Act keeps tightening: obligations for high-risk AI systems became fully applicable on August 2, 2026. For a European SME using ChatGPT, Claude or Gemini every day, this regulatory split deserves a clear-headed reading before drawing the wrong conclusions.
In brief
- On September 2, 2026, all twenty G20 nations endorsed the Carolina Principles, a non-binding text proposed by the US administration in Chapel Hill, North Carolina.
- The text encourages governments to reserve new AI rules for genuinely novel cases, avoid creating new regulators, and favor close collaboration with industry.
- The European Union pushed back: Commissioner Henna Virkkunen noted that 80% of Europeans consider careful AI regulation important for safety, citing a Eurobarometer survey.
- This G20 framework does not replace or loosen the AI Act: an SME established or active in the European Union remains fully subject to EU obligations, regardless of which AI tools it uses.
- Formal adoption is planned at the G20 leaders' summit in December 2026, in Florida.
What do the Carolina Principles say?
The Carolina Principles are a non-binding AI governance text, proposed by the United States and endorsed by every G20 member (including China and Russia) as well as the African Union. It was negotiated at an innovation ministerial summit held on September 1-2, 2026, in Chapel Hill, North Carolina, hence the name.
Three commitments summarize its spirit:
- Reserve new regulation for genuinely novel issues, rather than treating every new AI capability as a problem requiring its own dedicated law.
- Favor a sector-specific approach (healthcare, finance, transport) over a single cross-cutting text covering every use of AI.
- Work closely with industry to assess emerging technologies, rather than creating new dedicated regulatory agencies.
The text is part of a broader ministerial statement recognizing the potential of emerging technologies for economic prosperity. It carries no fines, no compliance obligation, and no deadline: it is a political statement of intent, not a treaty.
September 1-2, 2026
Chapel Hill summit
August 2, 2026
AI Act: high-risk obligations
December 2026
G20 leaders' summit
An openly assumed fault line with Europe
Europe's reaction was swift. Henna Virkkunen, the EU Commissioner for digital affairs, publicly rejected the idea that regulation and innovation are opposites. She pointed to a Eurobarometer survey showing that 80% of Europeans consider careful AI regulation important for safety. According to Axios, she also argued that the EU and the US often converge on similar protections, with Europe acting through advance regulation and the US relying more on courts and state laws.
At the same time, the European Commission sent information requests to more than thirty AI companies, as part of preliminary checks under the AI Act, covering safety and copyright issues. The message is clear: the European agenda is not slowing down just because Washington is choosing a different path.
G20 / US approach
Non-binding framework, no fines, no timeline. Priority on cooperating with industry. New rules reserved for genuinely novel cases. Sector-specific rather than cross-cutting approach.
EU approach (AI Act)
Binding regulation, with fines up to 15 million euros or 3% of global annual turnover. Cross-cutting obligations by risk tier. Precise enforcement timeline, already in force since August 2024 for the first milestones.
What this actually changes for an SME
For an SME established in the European Union, or selling products and services there, nothing changes immediately. The Carolina Principles are a US domestic policy orientation and a non-binding multilateral stance: they carry no legal weight in Europe and suspend none of the AI Act's obligations.
Still, three points deserve a business owner's attention:
- Regulatory uncertainty is here to stay. Two major economic blocs now display openly opposing philosophies. An SME exporting to the US, or opening a subsidiary there, may eventually face lighter local rules, without that exempting it from its EU obligations on its home market.
- AI providers (OpenAI, Anthropic, Google) remain subject to the AI Act as soon as they operate in Europe, regardless of nationality. Using a US-made tool does not exempt a European SME from checking that provider's compliance.
- Compliance pressure is not easing. The information requests sent by the European Commission to more than thirty AI companies show that AI Act enforcement is intensifying, independent of the international context.
Comparison table
| Criterion | Carolina Principles (G20) | AI Act (European Union) |
|---|---|---|
| Nature | Non-binding political statement | Legally binding regulation |
| Fines | None | Up to €15M or 3% of global turnover |
| Scope | Sector-specific recommendation, case by case | Cross-cutting obligations by risk tier |
| Timeline | Formal adoption targeted for December 2026 | High-risk obligations applicable since August 2026 |
| Effect for an EU SME | No direct legal effect | Applicable as soon as a high-risk AI system or a chatbot is used |
FAQ
Do the Carolina Principles replace the AI Act for a European company?
No. The Carolina Principles are a G20 text, non-binding, with no legal standing under EU law. An SME established in the European Union, or selling into it, remains fully subject to the AI Act.
Why did the United States propose this framework?
According to Reuters and several business outlets, the US administration wants to avoid a proliferation of AI-specific regulations worldwide, fearing they would slow innovation and the commercial spread of American technologies.
Does using ChatGPT, Claude or Gemini expose my SME to a compliance risk?
The tool itself is not the issue: these providers are subject to the AI Act as soon as they operate in the European market. Compliance risk for an SME mainly depends on how it uses the tool (customer chatbot, automated decision, generated content) and on meeting transparency obligations already in force, such as those under Article 50 of the AI Act.
Will this international divergence keep growing?
Likely, in the short term. Formal adoption of the Carolina Principles is planned at the G20 leaders' summit in December 2026, while the AI Act follows its own separate timeline. SMEs active internationally should track both agendas separately rather than wait for convergence.
In summary
The G20 has just displayed, more clearly than ever, two competing visions of AI governance: a light, sector-based American approach, and a binding, cross-cutting European one. For an SME, the key takeaway fits in one sentence: what gets decided in Chapel Hill or Doral changes nothing about what already applies in Brussels. Staying informed on both fronts remains an advantage, especially for companies that export or plan a US presence.
To dig deeper into AI Act deadlines and their practical implications, check out our other AI resources for business.
Sources: The White House, G20 Innovation Ministerial statement (whitehouse.gov, September 2, 2026); Techmeme citing Bloomberg; Axios, statements by Henna Virkkunen, EU Commissioner for digital affairs; European Commission, digital-strategy.ec.europa.eu.
