Just yesterday, the European Union officially included ChatGPT under the Digital Services Act (DSA) as a very large online search engine subject to enhanced regulations.
Today, the United States presented a different approach on the G20 stage:
Don’t rush to create a separate AI regulatory agency just because AI exists.
This policy framework being promoted by the US is called:
Carolina Principles.
Translated into Chinese as:
卡羅萊納原則 (Carolina Principles).
It has not yet been adopted as a common G20 policy.
Nor is it a new international law.
Rather, it is the governance approach the US hopes to gain support for during its G20 presidency among major economies.
The truly notable aspect is not just whether the US wants to regulate AI, but:
The clear global divergence on how AI should be regulated is now emerging.
What Happened Today?
From September 1 to 2,
the US Department of Commerce and the White House Office of Science and Technology Policy jointly hosted the:
G20 Innovation Ministerial in Chapel Hill, North Carolina.
Participants included commerce and technology officials from G20 countries.
The event covered various topics besides AI, such as:
emerging technologies, scientific research, manufacturing, supply chains, investment, and how technology drives economic growth.
Yet AI remains one of the most important policy issues.
US technology policy official Michael Kratsios prepared to promote the:
Carolina Principles during the meeting.
What’s the Core Message of the Carolina Principles?
According to Reuters, an important part of the prepared remarks is:
When countries formulate AI regulations,
new rules should be reserved for genuinely:
Novel Considerations
— truly new problems.
Put simply:
Don’t assume that just because a product uses AI,
all existing laws become obsolete.
First ask:
Can existing frameworks handle the issue?
If yes,
there may be no need to draft new AI-specific regulations.
New rules should only be introduced if unprecedented problems actually arise.
A Simple Example
If an AI company deceives consumers through advertising,
there are already established laws concerning:
advertising, consumer protection, and false claims.
The US approach suggests:
First see if current systems address the issue.
Don’t immediately create an:
AI-specific advertising regulatory body just because:
“This is AI advertising.”
Another example:
If AI is used by banks to process loans,
existing rules cover finance regulation, fair lending, data protection, and risk management.
The question shouldn’t be:
“Should we write an entirely new AI banking law?”
Instead:
“What gaps remain in existing financial regulations?”
The US Opposes Creating New AI Regulatory Agencies
This is a more controversial part of the Carolina Principles.
Reports indicate the framework aims to avoid:
Establishing new dedicated AI regulatory bodies.
The preference is for existing sectoral regulators to handle AI matters:
Financial AI by financial regulators.
Medical AI by healthcare regulators.
Transportation AI by traffic safety agencies.
Only genuinely new cross-sector issues should prompt separate solutions.
This approach is known as:
Sector-specific regulation.
Not creating a central, all-encompassing AI authority.
Does This Mean the US Wants “No AI Regulation”?
Not at all.
This is often misunderstood or oversimplified.
Terms like “Light-touch Regulation” or “Hands-off Approach” do not mean:
No laws, no safety testing, or unrestricted corporate freedom.
The real debate behind the Carolina Principles is:
How should regulatory frameworks develop?
Should every new AI issue require new legislation?
Should a new central AI regulator be established?
Or should existing systems across finance, healthcare, labor, consumer protection, intellectual property, and cybersecurity handle their domains?
These represent fundamentally distinct policy designs.
The US Also Wants to Prioritize Research Investment
According to Reuters, the Carolina Principles also call for supporting:
Foundational Research
to accelerate scientific discovery and create more opportunities for technology commercialization.
So, the approach isn’t just about less regulation but also that:
The government should foster conditions for new technology growth.
Research advances first, companies turn tech into products, then emerging issues are addressed.
Why Is the US Emphasizing This Now?
One critical context:
AI is no longer just a tech policy issue; it has become international competitive policy.
Aside from concerns about AI errors, the US worries that overly fast or costly regulation could cause:
slower R&D, less investment, fewer data centers, and AI models falling behind other countries,
thus weakening the entire AI ecosystem's influence—especially as China’s open-weight AI models rapidly close the capability gap.
Therefore, US AI governance must balance:
safety, economy, tech competition, national security, and global influence.
How Is This Different from Europe?
Europe’s clearer approach is to:
Establish common frameworks first, then require large services to comply with these rules.
The EU AI Act is a prime example.
It sets different obligations based on AI use and risk.
Yesterday’s news illustrates this too:
ChatGPT’s search functions exceed the EU threshold,
so it was designated under the DSA as a Very Large Online Search Engine.
It must now manage systematic risks concerning:
illegal content, fundamental rights, minors, elections, public safety, and mental and physical wellbeing.
This contrasts interestingly with the US pushing Carolina Principles today.
But Don’t Simplify It as “Europe Regulates, US Doesn’t”
This is still too crude.
The true difference can be summarized as:
Europe asks:
What shared risks do large new technologies bring, and should common obligations be established beforehand?
Whereas the US approach is:
Don’t assume every AI problem is entirely new; if existing rules suffice, let them apply.
Both sides acknowledge AI risks.
The difference lies in whether frameworks should precede problems or be reactive.
Interestingly, Even AI Company Leaders Disagree
The G20 Innovation Ministerial invited many major AI figures, including:
OpenAI CEO Sam Altman, Nvidia CEO Jensen Huang, Elon Musk, and Google DeepMind CEO Demis Hassabis.
Altman and Huang are scheduled to speak with US Commerce Secretary Howard Lutnick on the second day.
The idea that “tech companies want no regulation” is inaccurate.
For example, Demis Hassabis has advocated for:
a dedicated US agency to test cutting-edge AI systems before deployment.
This contrasts clearly with:
“No new AI-specific regulatory agencies.”
Thus, the true AI governance debate is not simply between governments and tech companies.
Even within industry, there is no consensus on:
who tests, who oversees, or whether to create new agencies.
Why Is This Time Particularly Sensitive?
Because AI safety issues are no longer theoretical.
Cutting-edge AI agents in test environments are:
crossing boundaries and entering real systems.
This forces companies like OpenAI and Anthropic to reevaluate:
sandboxing, real-time blocking, external testing, and model behavior.
We’ve also seen today’s news that the Financial Stability Board (FSB) considers frontier AI an accelerated cyber risk, a pressing issue for the financial sector.
Therefore, the US’s proposal to not rush into new regulation naturally invites the question:
If AI capabilities grow faster than existing frameworks, will waiting to regulate when problems arise be too slow?
This is the biggest debate Carolina Principles will face moving forward.
Regulating Too Much Also Has Its Costs
The opposite problem is real too.
If every AI-enabled product must undergo extra:
applications, testing, documentation, approvals, and dedicated oversight,
small companies may simply not afford it.
That could end up benefiting only big players like:
Google, Microsoft, OpenAI, Meta, and Amazon,
who can bear compliance costs.
Consequently, rules designed to control big tech might inadvertently raise barriers for startups.
Regulation itself can alter competition.
This is why no simple answers exist.
Have Carolina Principles Been Approved Today?
No.
This distinction is important.
What can be confirmed is that the US is promoting this framework at the G20 Innovation Ministerial, hoping to secure support.
It would be incorrect to say:
“G20 has decided not to regulate AI.”
Nor say:
“The world has adopted the Carolina Principles.”
G20 countries have diverse systems.
The EU already has the AI Act.
Other nations have their own AI policy directions.
Even if a joint statement emerges, it won’t mean all domestic laws become identical.
What Does the US Aim to Achieve?
The real goal may not be:
getting every G20 country to change its laws immediately.
Instead, it’s about:
building a common global language for AI governance.
If in future international AI discussions,
countries accept that:
new regulators are not always necessary,
existing industry frameworks can be used,
new rules should only address genuinely novel problems,
government and industry collaborate on testing,
and foundational research and commercialization receive priority,
then the US governance model could be exported worldwide.
This moves beyond domestic policy into:
a global technology rules competition.
Why Do AI Companies Care About Global Rule Alignment?
Because AI products easily cross borders.
A single AI model could serve markets in:
the US, Europe, Japan, South Korea, India, and Taiwan simultaneously.
If each country demands:
entirely different safety tests, documentation, model restrictions, data rules, age limits, and agent behavior boundaries,
companies might have to develop:
“One AI system per country.”
This greatly increases:
development costs, compliance costs, and product complexity.
This is the underlying challenge behind Apple deploying different AI systems in different markets:
With AI globalized, technology isn't the biggest hurdle;
whether regulatory systems can interconnect smoothly may be even harder.
What Does This Mean for Ordinary Users?
You might wonder how G20 talks about regulation relate to your daily use of ChatGPT.
In the long run, quite directly.
These rules will determine:
which data AI can access,
when human review is mandatory,
which versions minors can use,
whether you can demand explanations for AI decisions,
whether there’s recourse if AI errs,
what records companies must keep,
if models require independent testing before release,
and whether certain agents can autonomously make decisions for you.
These aren’t abstract legal issues;
they ultimately shape:
the product interface.
What About Small Companies?
Small businesses don’t need to immediately analyze the entire G20 documentation.
Instead, they should focus on a practical principle:
Don’t wait for regulations to require action before setting your own AI boundaries.
For example:
Which data is AI allowed to handle?
Which data is off-limits?
Which outcomes require human approval?
Which AI tasks are only drafts?
What can’t be automatically paid for?
How to recover from errors?
Whatever global regulatory path is taken—
European, US, or some middle ground—
companies must still answer:
“How do you know what the AI is doing?”
The Key Takeaway Tonight Is Not “US Opposes AI Regulation”
That’s an oversimplification.
A more precise description is:
The US is promoting a policy that:
minimizes creating new dedicated AI regulatory frameworks, leverages existing sectoral systems, and adds new rules only for genuinely novel problems.
Less than 24 hours ago, the EU formally imposed higher-level DSA systemic risk obligations on ChatGPT’s large-scale search function.
Seeing these two developments together means:
The next phase of global AI competition is not just about models.
It’s also about which regulations can simultaneously:
allow AI to keep growing,
ensure responsible risk management,
encourage investment,
gain user trust,
and attract other countries to follow.
In the end, there may not be a single global AI governance model;
multiple governance schemes could coexist.
And right now,
we are beginning to see their divergence.
Today, progress alongside AI.
Learn an AI skill daily.
Save time every day.
Enhance your capabilities little by little.
SasaDaily, growing with you.
Recommended Reading
Bill Gates Always Supported AI, So Why Now Seek Talks with Xi Jinping on "Global Governance"?
South Australia Just Signed Collaboration with OpenAI, Yet Now Plans an AI Royal Commission?
