⏱️ Lectura: 10 min
“We don’t want a DMV for artificial intelligence”: with that phrase, US officials summed up their rejection of centralized regulatory oversight of AI regulation at an international technology governance forum this week.
📑 En este artículo
Days later, China called for cooperation instead of rivalry in artificial intelligence, just ahead of a summit between Xi Jinping and Donald Trump. The contrast exposes two opposing views on how to govern a technology that already moves billions of dollars in public investment.
TL;DR
- US officials used the phrase “no DMV for AI” to reject a body that would license models before deployment, according to CNET.
- China called for cooperation in AI, not rivalry, ahead of the summit between Xi Jinping and Donald Trump, according to South China Morning Post.
- South Korea requested a record $21.6 billion science budget for 2027, according to UPI, as Asia accelerates public spending on AI.
- The European Union already applies the AI Act, the first comprehensive framework that classifies AI systems by risk level, in effect since 2024.
- The United States favors voluntary guidelines like the NIST AI RMF over the European model of mandatory audits.
- China combines state registration of algorithms with an international discourse of openness and cooperation.
- The Xi-Trump summit is shaping up to be the next friction point over AI governance.
Introduction
The question of who regulates artificial intelligence, and under what rules, is no longer an academic debate. In recent months, governments in the United States, China, the European Union, and South Korea have moved different pieces on the board: some are cutting rules, others are calling for cooperation, and others are simply allocating more public money. AI regulation has become the ground where much of 2026’s tech geopolitical competition is being played out.
What happened with AI regulation
At an international forum on artificial intelligence governance, US officials defended a hands-off stance: they rejected the idea of creating a body that would act as a DMV for AI, referring to the Department of Motor Vehicles, the state agency that issues driver’s licenses in the US and is synonymous with slow, bureaucratic red tape. The comparison aimed to convey that licensing or auditing every AI model before deployment would stifle innovation, according to CNET.
Almost simultaneously, China took a different stance in the same debate. According to the South China Morning Post, Beijing called for international cooperation on AI instead of a competitive race, a calculated message just ahead of a planned summit between Chinese President Xi Jinping and US President Donald Trump.
Context and history of AI regulation
The clash isn’t new, but it has intensified. Since 2023, different blocs have tried opposing approaches to AI regulation. The European Union passed the AI Act, the first comprehensive legal framework that classifies AI systems by risk level and requires audits for “high-risk” ones. The United Kingdom opted for one-off safety summits, the series that started at Bletchley Park in 2023. The United States, meanwhile, moved forward with executive orders and voluntary NIST guidelines, without a single federal law requiring licenses.
China developed its own framework: it requires registration of recommendation algorithms and specific rules for AI-generated content, but combines that state oversight with an international discourse of openness, like the one it repeated now ahead of the summit with Trump. The rise in science budgets across Asia reinforces that reading: South Korea requested a record $21.6 billion budget for 2027, according to UPI, much of it earmarked for science and AI infrastructure.
Technical details: how AI regulation gets audited
Behind the phrase “no DMV for AI” lies a concrete technical discussion: what mechanisms to use to verify that an AI model is safe before or after deployment. Regulatory frameworks that do require oversight, like the European AI Act, rely on three main technical pieces.
- Model cards: structured documents that describe a model’s training data, known limitations, and intended use cases.
- Third-party audits: red-teaming evaluations and adversarial testing before deployment in high-risk sectors.
- Incident registries: databases where failures or harms caused by AI systems in production are reported.
The United States doesn’t reject these practices in themselves: the NIST AI Risk Management Framework recommends them voluntarily. What it rejects is turning them into a mandatory, centralized process, equivalent to requiring a license before operating, as is the case with the AI Act in the European Union.
💡 Tip: documenting a model card doesn’t depend on it being mandatory: it serves as evidence if a regulator asks for explanations after an incident.
How to start documenting your AI
If your company deploys AI models and you want to get ahead of any of these frameworks, the technical starting point is the same in almost every country: document the model with a model card and keep your own incident registry, even if it isn’t mandatory yet where you operate.
A minimal model card, in YAML format, looks like this:
model_card:
name: "recomendador-catalogo-v3"
version: "3.2.0"
entrenado_con: "internal catalog 2024-2026, 4.2M items"
uso_previsto: "product ranking in internal search"
limitaciones_conocidas:
- "bias toward items with more reviews"
- "not evaluated for catalogs under 10k items"
contacto_responsable: "[email protected]"And an incident registry, in JSON, that you can version alongside the model in the same repository:
{
"incident_id": "INC-2026-014",
"model": "recomendador-catalogo-v3",
"fecha": "2026-08-30",
"severidad": "medium",
"descripcion": "the model repeated discontinued recommendations",
"mitigacion": "stock filter added before serving results",
"estado": "resolved"
}To confirm which rules apply to you today depending on where you operate, the most direct way is to check the OECD.AI policy tracker, which brings together regulatory initiatives from dozens of countries, and compare it against the official AI Act summary if you operate in the European Union.
Impact and analysis of AI regulation
The practical result is a fragmented regulatory map that AI companies with global operations already have to navigate.
| Approach | Example | Advantage | Limitation |
|---|---|---|---|
| Comprehensive risk-based law | AI Act (European Union) | Clear, enforceable rules by sector | High compliance costs for startups |
| Voluntary guidelines + sector-specific | NIST AI RMF (US) | Doesn’t slow the pace of model releases | No real penalty for noncompliance |
| State registration + openness discourse | Chinese algorithm framework | Direct control over local deployment | Little externally verifiable transparency |
| Multilateral principles | OECD, AI safety summits | Coordination between different blocs | Not legally binding |
None of the four models solved the underlying problem: how to audit a system that changes its behavior depending on the data it receives, without slowing down its development. The US position bets that the market and reputation will correct errors faster than a regulator; the European one bets that an avoided incident is worth more than lost speed.
flowchart TD
A["Global AI company"] --> B["Operates in the US"]
A --> C["Operates in the European Union"]
A --> D["Operates in China"]
B --> E["NIST AI RMF: voluntary guidance"]
C --> F["AI Act: mandatory audit"]
D --> G["State algorithm registry"]
subgraph COORD["International coordination"]
H["OECD: non-binding principles"]
end
E --> H
F --> H
G --> H📌 Note: China’s cooperation message arrived just before the summit with Trump, a timing that analysts read as a negotiation signal, not just domestic politics.
No position is risk-free. An approach without mandatory licenses, like the one the US advocates, leaves it up to each company to decide when a model is safe enough for production, without an externally verifiable minimum standard. A licensing approach like the European one, on the other hand, can take months to approve a high-risk system, time that a startup with limited funding doesn’t always have.
What’s next for AI regulation
The summit between Xi Jinping and Donald Trump is the next point where this contrast will be tested in public. Whether China’s cooperative tone translates into concrete proposals, like shared technical standards or agreements on incident reporting, or remains diplomatic rhetoric, is the open question.
At the same time, the rise in science budgets across Asia, like the record $21.6 billion request South Korea made for 2027, suggests that much of the race is no longer being played out only in laws, but in how much public money each country is willing to put behind its own AI infrastructure.
📖 Summary on Telegram: View summary
If you want to see your country’s up-to-date regulatory status today, check the OECD.AI policy tracker and look up your jurisdiction before launching your next model into production.
Frequently Asked Questions
What does “no DMV for AI” mean?
It’s the phrase US officials used to reject the idea of a centralized body that would license AI models before deployment, comparing it to the bureaucratic process of getting a driver’s license at the Department of Motor Vehicles.
What is the European Union’s AI Act?
It’s the first comprehensive legal framework that classifies AI systems by risk level and requires mandatory audits for “high-risk” ones, in effect in the European Union since 2024.
Why did China call for cooperation right before the summit with Trump?
The message aims to position China as an actor open to coordinating global AI rules at a moment of high-profile bilateral negotiation with the United States.
Does the United States not regulate AI at all?
It does regulate it, but on a sector-by-sector basis and largely on a voluntary footing, supported by guidelines like the NIST AI Risk Management Framework, without a single federal law requiring licenses.
What is a model card?
It’s a structured document that describes an AI model’s training data, known limitations, and intended use, used as evidence of transparency for regulators or users.
How does this affect a company that doesn’t operate in the US or the EU?
It’s still worth checking the local framework: more and more countries, including several in Asia and Latin America, are drafting their own rules based on the European AI Act or the OECD’s voluntary principles.
References
- CNET via MSN: report on the US stance against centralized AI regulation at a global forum.
- South China Morning Post via MSN: coverage of China’s call for AI cooperation ahead of the Xi-Trump summit.
- UPI: record $21.6 billion budget requested by South Korea’s science ministry for 2027.
- NIST AI Risk Management Framework: voluntary AI risk management guidance used as a reference in the United States.
- AI Act (European Union): official information on the European AI regulatory framework by risk level.
- OECD.AI: policy tracker and AI principles adopted by OECD member and partner countries.
📱 Enjoy this content? Follow @programacion on Telegram for daily tech content in Spanish: quick summaries, fresh content every day.
Imagen destacada: Foto de Andy Feliciotti en Unsplash
0 Comments