---
title: "Why AI Regulation Differs by Country — Three Paths Rooted in Political Systems"
url: https://blog.tyrano.dev/en/why-ai-regulation-differs-by-country-three-paths-rooted-in-political-systems
lang: en
site: Tech AI News - 테카이
category: "AI Industry Analysis"
tags: ["ai","ai-regulation","eu-ai-act","us-ai-policy","china-ai-policy","governance"]
published: 2026-10-26T07:11:24.630Z
updated: 2026-10-27T12:34:56.820Z
sources:
  - https://www.euronews.com/2026/10/07/laws-pledges-or-an-iron-grip-how-the-eu-us-and-china-are-reining-in-ai
  - https://cadeproject.org/updates/eus-digital-omnibus-proposes-major-delay-to-ai-act-high-risk-rules/
  - https://digitalpolicyalert.org/event/38517-cyberspace-administration-of-china-announced-registered-generative-ai-service-information-for-january-to-february-2026
  - https://www.mondaq.com/unitedstates/new-technology/1740258/us-artificial-intelligence-law-update-navigating-the-evolving-state-and-federal-regulatory-landscape
  - https://mobile.newsis.com/view/NISX20241226_0003011314
---

# Why AI Regulation Differs by Country — Three Paths Rooted in Political Systems

> Facing the same AI technology, Europe legislates, the US exchanges pledges with companies, and China manages it through registration. This difference isn't because the three countries have different philosophies about AI — it's because the way each one makes laws in the first place is designed differently.

---

## Background — Why This Topic, Why Now

Heading into 2026, AI regulation in the US, EU, and China appears to be diverging into increasingly different shapes on the surface. The EU keeps pushing back the enforcement date for high-risk systems under the AI Act it adopted in 2024, while keeping the law's overall framework intact. The US, lacking federal legislation, is filling the gap with executive orders and voluntary corporate pledges. China continues to expand a system that requires prior registration for every generative AI service.

Framing these three paths as a question of "who trusts or fears AI more" offers a thin explanation. As [Euronews](https://www.euronews.com/2026/10/07/laws-pledges-or-an-iron-grip-how-the-eu-us-and-china-are-reining-in-ai) reports, all three regions worry equally about risks like child safety, cyberattacks, and disinformation. The reason their responses differ so much instead comes down to how each country has already decided to make new rules and who enforces them — differences in legislative structure and the distribution of administrative power. AI is simply the latest case being slotted into that existing framework.

## Key Data & Current State

- **Delayed enforcement of the EU's high-risk rules:** The high-risk system obligations at the core of the EU AI Act were originally set to take effect in August 2026. Following Digital Omnibus negotiations, standalone high-risk systems (Annex III) were pushed to December 2027, and high-risk systems embedded in products (Annex I) to August 2028 ([CADE Project](https://cadeproject.org/updates/eus-digital-omnibus-proposes-major-delay-to-ai-act-high-risk-rules/), 2026).
> Interpretation: This isn't a reversal of the law itself — just an adjustment of the enforcement date to match when the standards and guidelines will be ready. Because changes require agreement among the three bodies — the European Parliament, Council, and Commission — delay is far easier than repeal.
- **China's generative AI registration tally:** According to China's Cyberspace Administration (CAC), a cumulative 796 generative AI services had completed registration (beian) by the end of February 2026 ([Digital Policy Alert](https://digitalpolicyalert.org/event/38517-cyberspace-administration-of-china-announced-registered-generative-ai-service-information-for-january-to-february-2026), 2026).
> Interpretation: This count is possible precisely because the system requires prior approval before a service can launch. The US and EU, by contrast, have no concept of a "number of approved AI services" at all.
- **The US federal-state legislative gap:** Since the Biden administration's AI executive order was rescinded in January 2025, the US has had no comprehensive federal AI law, leaving individual states such as Colorado and California to fill the gap with their own legislation ([Mondaq](https://www.mondaq.com/unitedstates/new-technology/1740258/us-artificial-intelligence-law-update-navigating-the-evolving-state-and-federal-regulatory-landscape), 2026).
> Interpretation: The very fact that 50 individual states can each write their own rules while Congress stays gridlocked is a phenomenon that only a federal system could produce.

## In-Depth Analysis

### The EU — Comprehensive Legislation First, Then Split Enforcement

The EU AI Act is a single regulation that applies simultaneously across 27 member states, but actual enforcement is split between each country's "national competent authorities" and the central AI Office. This is possible because the EU had already tested the "comprehensive legislation plus distributed enforcement" model with the GDPR. Having the European Parliament and Council agree on the text of the law first, then filling in the details later through implementing acts and guidelines, is the standard procedure of the EU's own legislative process. That's why the AI Act, too, ends up being negotiated not over "whether to pass it" but over "when it should take effect."

### The US — Fitting AI into Existing Agencies

Because of the US constitutional structure — the division of power between federal and state governments (federalism) and the separation of the executive and legislative branches — rather than creating a new agency just for AI, existing sector-specific regulators like the FTC, FDA, and EEOC each handle AI within their own jurisdiction. While Congress fails to agree on comprehensive AI legislation, the President fills the gap with executive orders and companies fill it with voluntary pledges to the White House. On September 29, 2026, the CEOs of six major AI companies signed a White House agreement pledging independent audits and board oversight, but it carries no penalty provisions ([Euronews](https://www.euronews.com/2026/10/07/laws-pledges-or-an-iron-grip-how-the-eu-us-and-china-are-reining-in-ai), September 2026). Analysis: the very fact that such voluntary pledges are possible is itself the flip side of federal legislation being stalled.

### China — Expanding an Existing Control Network

China's generative AI regulation isn't an entirely new framework — it's the result of fitting AI services into the internet content and algorithmic recommendation control network that the CAC was already operating. The Interim Measures for the Management of Generative AI Services, which took effect in August 2023, formalized that framework, and the scope of what must be registered has kept expanding since. Because the executive branch can draft and enforce detailed rules (ministerial regulations) quickly without prior legislative approval, when a new risk emerges — like companion chatbots — China can add a dedicated regulation within months (one took effect in July 2026).

## Implications for Korea

South Korea's Basic AI Act was designed to take initial effect on January 22, 2026, with provisions requiring subordinate legislation applying additionally from July 21 of the same year ([Newsis](https://mobile.newsis.com/view/NISX20241226_0003011314), December 2024). In passing comprehensive legislation first and filling in the details through implementing decrees, Korea is closer to the EU model than the US one. That said, the Ministry of Science and ICT's announcement that it will defer enforcement for at least a year to help the industry adjust follows the same pattern as the EU delaying its high-risk enforcement date — the law comes first, and the intensity of enforcement is adjusted later.

## Outlook & Variables to Watch

- **Positive variable:** All three models share the trait that "rules follow only after a risk becomes concrete," leaving room for the system to be refined as real-world harm cases accumulate.
- **Risk variable:** If the rules across the three blocs remain incompatible, companies launching services across multiple markets simultaneously will have to pay the cost of either designing to the "strictest standard" or building region-specific versions.
- **Checkpoints:** Whether the EU's high-risk rules actually take effect (2027–2028), whether the US Congress resumes legislating after the November midterm elections, and when Korea's deferral of the Basic AI Act ends are the next inflection points.

## Conclusion

1. The differences in the shape of AI regulation are a direct reflection of differences in how political systems are designed — none of the EU's comprehensive legislation, the US's distributed enforcement, or China's prior-approval system was created because of AI.
2. Rather than converging, the three models are each adjusting their pace along their own trajectory: the EU delays its enforcement dates, the US fills gaps with voluntary pledges, and China keeps widening its control network.
3. Countries like Korea that have chosen the comprehensive-legislation model should keep an eye on the EU-style timing pattern of "pass the law, defer the enforcement."
