AI regulation in Asia-Pacific: China, South Korea, Japan and Singapore in 2026
We already compared the EU's AI Act with the US regulatory vacuum. Asia-Pacific doesn't offer a single third model — it's four distinct approaches coexisting in the same region, from mandatory labeling in China to the total absence of penalties in Japan.
China — mandatory labeling since 2023, and now content too
China was among the first countries in the world to regulate generative AI, with its Interim Measures for the Management of Generative AI Services, in force since August 2023. Since then, regulation has grown more granular: as of September 1, 2025, the Measures for Labeling of AI-Generated Content require explicit (user-visible) and implicit (embedded metadata) markings on all synthetic content — text, image, audio and video generated or substantially edited by AI — published on Chinese platforms. Alongside AI Act Art. 50, it's one of the strictest AI content-labeling rules already in force anywhere in the world.
South Korea — the most recent comprehensive law, with real penalties
South Korea's AI Basic Act took effect on January 22, 2026, placing the country alongside the EU as one of the jurisdictions with the most developed comprehensive AI framework. It classifies systems into three tiers: generative AI (mandatory content labeling), high-impact AI (critical sectors like healthcare and transportation, with fundamental-rights impact assessments and human oversight), and high-performance AI (models trained with 10²⁶ FLOPs or more, subject to lifecycle risk-management plans). Administrative fines reach up to 30 million KRW (roughly $21,000), with a one-year grace period before penalties are actually enforced.
Japan — the clearest bet on soft law
At the opposite extreme is Japan. Its AI Promotion Act, passed on May 28, 2025, is the country's first comprehensive AI legislation, but explicitly prioritizes innovation over restriction: it imposes "reasonable effort" obligations on AI developers, providers, and business users, and requires cooperation with the government's AI Strategic Headquarters — but includes no monetary penalty for non-compliance. It's the clearest example of a regulatory model built on guidelines and cooperation rather than sanctions.
Singapore — the world's first framework specifically for agentic AI
Singapore has positioned itself as a reference point for voluntary governance and regulatory anticipation. On January 22, 2026, at the World Economic Forum, it unveiled what is described as the world's first comprehensive governance framework specifically designed for agentic AI — systems capable of autonomous reasoning, planning, and action. The framework isn't binding, but rests on four pillars (bounding risks upfront, ensuring meaningful human accountability, technical controls, and end-user responsibility) and is already shaping AI governance initiatives across ASEAN, while complementing existing tools such as AI Verify.
What this means for a company with regional presence
Four countries, four approaches: mandatory labeling and content control in China, a comprehensive law with penalties and precise technical thresholds in South Korea, a cooperation-based framework with no penalties in Japan, and cutting-edge voluntary guidelines in Singapore. For a company operating across the region, there's no single reference standard like the AI Act in the EU — the most efficient strategy is to treat risk classification and AI content labeling as baseline requirements applicable everywhere, and add jurisdiction-specific layers (penalties in Korea, content controls in China) depending on where each system is deployed.
A common reference framework for operating across countries
AI Act Starter Kit as a risk-classification baseline applicable to any jurisdiction, and the AI Frameworks Comparison Table to map which standard covers which requirement.