What is AI regulation in the Asia-Pacific?

AI regulation: countries and regions

AI regulation in the Asia-Pacific is not one regime but a spectrum. At one end, China enforces binding rules on algorithms, deep synthesis and generative AI. At the other, Singapore, Japan, Australia and New Zealand rely mainly on voluntary frameworks and existing law. South Korea, Taiwan and Vietnam have now passed comprehensive AI statutes, and a non-binding ASEAN Guide sets shared regional principles. The picture is shifting fast as of June 2026.

Reviewed by Jackie, Head of Learning & Development, Levellers · Last reviewed 8 June 2026

What this means

There is no single Asia-Pacific approach to AI. Governments have chosen different tools for different reasons, and they sit along a spectrum from hard, enforceable law to soft, voluntary guidance. China sits at the binding end, with mandatory rules and national standards that providers must follow before launching public services. Singapore, Japan, Australia and New Zealand sit nearer the voluntary end, leaning on guidelines, principles and the laws they already have.

The middle ground is filling in quickly. South Korea's comprehensive AI Basic Act took effect in January 2026, Taiwan passed an AI Basic Act in December 2025, and Vietnam's binding AI Law took effect in March 2026. India has chosen a light-touch, principles-based path. Above the national layer, the ASEAN Guide on AI Governance and Ethics gives ten Southeast Asian states a shared, voluntary reference.

This page orients you to the regional shape and routes you to the country detail. It does not restate each country in full. Use the links to reach the dedicated articles for China, Japan, South Korea, Singapore, Australia, New Zealand and others.

Why it matters

The Asia-Pacific includes some of the world's largest AI developers and markets, so the rules here shape global practice. A company operating across the region cannot assume one compliance model will travel. In China, a public-facing generative AI service may need security assessment, algorithm filing and content labelling before launch. In Singapore or Japan, the same product may face mainly voluntary expectations, backed by existing privacy, consumer and copyright law. Several regimes apply outside their borders, so a firm with no local office can still fall within scope. Getting the map right early avoids both over-engineering compliance in light-touch markets and under-preparing for the binding ones.

How it works

The spectrum from binding to voluntary

The clearest way to read the region is as a spectrum. China anchors the binding end with enforceable measures and mandatory national standards. South Korea, Taiwan and Vietnam have moved to comprehensive statutes. Japan, Singapore, Australia, New Zealand and India sit nearer the voluntary end, relying on guidance plus existing law. The ASEAN Guide adds a shared, non-binding regional layer. Positions are forming, and the gap between the binding and voluntary ends has narrowed over 2025 and into 2026.

China: binding rules and national standards

China has the region's most developed binding regime. Three core instruments stack together: the Provisions on the Administration of Algorithmic Recommendation in Internet Information Services, effective March 2022, which created the algorithm registry; the Provisions on the Administration of Deep Synthesis, effective January 2023, which target synthetic media; and the Interim Measures for the Management of Generative AI Services, effective 15 August 2023, China's first binding generative AI rules. Public-facing services with public opinion or social mobilisation properties must complete security assessment and algorithm filing. Mandatory AI content labelling rules took effect on 1 September 2025. The TC260 AI Safety Governance Framework, updated to version 2.0 in September 2025, is a technical reference that feeds binding national standards. See the China article for detail.

South Korea, Taiwan and Vietnam: comprehensive statutes

South Korea's AI Basic Act took effect on 22 January 2026, making it one of the first comprehensive national AI laws, with obligations on high-impact AI and generative AI transparency. The Ministry of Science and ICT is operating a grace period of at least one year in 2026 deferring fines; administrative fines reach up to KRW 30 million (about US$20,700) for failures such as non-notification of AI use or failure to appoint a domestic representative. High-impact AI, including hiring tools, requires impact assessments, explainability, human oversight and documentation; separately, frontier high-performance models trained with at least 10 to the power of 26 FLOPs face safety obligations, a threshold set ten times higher than the EU AI Act. Taiwan passed its AI Basic Act in December 2025, a principles-based framework naming the National Science and Technology Council as lead. Vietnam's binding AI Law took effect on 1 March 2026 with a risk-based model and extraterritorial reach. See the country articles.

Japan: promotion-first, soft-law approach

Japan passed its AI Promotion Act in 2025, taking an innovation-first stance with no direct fines. The practical compliance layer is the AI Guidelines for Business, issued by METI and MIC, treated as a comply-or-explain standard. Existing laws on privacy, copyright and competition still apply. See the Japan article.

Singapore: model governance and testing tools

Singapore is the leading example of the voluntary, framework-driven model. IMDA's Model AI Governance Framework, first issued in 2019, was extended to generative AI in 2024 and to agentic AI in January 2026. AI Verify provides a testing and assurance toolkit. Compliance is voluntary, though organisations remain accountable under existing law. See the Singapore article.

Australia and New Zealand: existing law plus guidance

Australia published a Voluntary AI Safety Standard with ten guardrails and consulted on mandatory guardrails for high-risk AI, but the National AI Plan of December 2025 set aside those mandatory guardrails in favour of existing laws plus a new AI Safety Institute. New Zealand has confirmed it will not build a standalone AI Act, taking a proportionate, risk-based approach, with a Public Service AI Framework and the Algorithm Charter for the public sector. See the Australia and New Zealand articles.

The ASEAN regional layer

ASEAN issued its Guide on AI Governance and Ethics in February 2024, a voluntary framework for all ten member states built on seven guiding principles. It was expanded in January 2025 with a generative AI edition, and ASEAN adopted a Responsible AI Roadmap for 2025 to 2030 in March 2025. The Guide is not binding but increasingly shapes national policy. See the dedicated ASEAN Guide article.

Examples

A developer launching a public chatbot in mainland China must complete the large model filing, passing both provincial-level and national-level technical testing before public release. Filed services include DeepSeek and Baidu's Ernie Bot. The provider must also apply explicit and implicit labels to generated content under the rules that took effect on 1 September 2025. As of December 2025, 748 generative AI services had completed national-level filing, alongside 435 local-level registrations.

A multinational deploying an AI hiring tool that affects users in South Korea may fall within the AI Basic Act's high-impact category and face impact assessments, explainability, human oversight and documentation duties, even without a Korean office, because the law applies extraterritorially. A foreign operator without a Korean office must appoint a domestic representative if it meets one of three thresholds: prior-year total revenue of at least KRW 1 trillion (about US$681 million), AI-service revenue of at least KRW 10 billion (about US$6.9 million), or at least 1 million average daily domestic users over the preceding three months.

A Singapore firm adopting a generative AI assistant can use IMDA's Model AI Governance Framework for Generative AI and the AI Verify testing toolkit to structure governance, while remaining legally accountable under existing law. None of these steps is mandated by a standalone AI statute.

Common misunderstandings

"There is one Asia-Pacific AI law." There is not. The region runs from binding Chinese rules to voluntary Singaporean frameworks, with comprehensive statutes now in Korea, Taiwan and Vietnam.

"Voluntary means there are no rules." Even where AI-specific law is voluntary, existing privacy, consumer, copyright and competition laws still apply, and breaching them carries real penalties.

"China has a single AI law." China regulates through stacked instruments on algorithms, deep synthesis, generative AI and content labelling, plus technical national standards, rather than one act.

"Japan's law imposes fines." The AI Promotion Act creates no direct financial penalties and relies on guidance and existing law.

"Australia is introducing mandatory high-risk guardrails." It consulted on them, but the December 2025 National AI Plan set them aside in favour of existing law.

Risks and boundaries

This is a fast-dating field. Positions stated here are current as at June 2026 and several are still forming. Many statutes rely on implementing decrees and guidelines that are not yet finalised, so detail can change. Extraterritorial reach is common, so a firm without a local presence can still be in scope. This hub is an orientation, not legal advice or a country-by-country compliance manual. For binding obligations, dates and thresholds, rely on the official primary sources and the dedicated country articles, and take local advice. Where official sources are in Chinese, Korean, Japanese or Vietnamese, English versions are often translations and should be checked against the original.

What to do next

Start by mapping where you actually operate and where your AI systems reach users, because extraterritorial rules mean reach matters as much as location. Triage your markets along the binding-to-voluntary spectrum: treat China, South Korea, Taiwan and Vietnam as binding-compliance markets with filing, assessment, labelling or conformity duties; treat Japan, Singapore, Australia and New Zealand as guidance-plus-existing-law markets. Adopt a common internal governance baseline aligned with widely used frameworks so you can meet voluntary expectations and prepare for hard rules at once. Track the implementing decrees due across Korea and Vietnam, and watch ASEAN's Responsible AI Roadmap and the Digital Economy Framework Agreement for regional convergence. Re-check positions every few months given the pace of change.

Have a question or a suggestion, or want to understand how we research and review these guides? Read about our editorial standards and how to reach us.

FAQs

Which Asia-Pacific country has the strictest AI rules?

China has the most developed binding regime, with enforceable measures on algorithms, deep synthesis, generative AI and content labelling, plus mandatory national standards. Public-facing generative AI services face security assessment and filing before launch.

Does the Asia-Pacific have a single AI law like the EU AI Act?

No. There is no regional binding statute. The closest regional instrument is the voluntary ASEAN Guide on AI Governance and Ethics. National approaches differ widely.

Which countries have passed comprehensive AI laws?

South Korea's AI Basic Act took effect in January 2026, Taiwan passed an AI Basic Act in December 2025, and Vietnam's AI Law took effect in March 2026. These are the region's main comprehensive statutes.

Is Singapore's AI governance binding?

No. Singapore's Model AI Governance Framework and AI Verify toolkit are voluntary. Organisations remain accountable under existing laws such as data protection.

Did Australia introduce mandatory AI guardrails?

No. Australia published a Voluntary AI Safety Standard and consulted on mandatory guardrails, but the December 2025 National AI Plan set them aside in favour of existing laws and a new AI Safety Institute.

Do these rules apply to companies based outside the region?

Often yes. Several regimes, including South Korea's and Vietnam's, apply extraterritorially where AI systems affect local users, and may require a local representative.

What does the ASEAN Guide cover?

It is a voluntary framework with seven guiding principles, issued in 2024 and expanded for generative AI in 2025. It shapes national policy across Southeast Asia but is not legally binding.

Sources

‹ What is AI regulation in the Americas?

What is AI regulation in the Middle East? ›