AI regulation in Asia-Pacific: what each country requires in 2026

The three models at a glance

• General, mandatory law: South Korea.
• Mandatory rules by type of service: China (recommendation algorithms, deep synthesis, generative AI and content labelling).
• Promotion without penalties: Japan.
• Voluntary guidance plus sector and data rules: Singapore, India and Australia.

South Korea: the region’s first general AI law

The Framework Act on AI (known as the AI Basic Act) has been in force since January 22, 2026. It is the closest to the EU AI Act, though shorter. The essentials:

• High-impact AI: systems that can significantly affect people’s lives or safety (for example hiring, credit decisions, healthcare or transport). They require risk assessment, lifecycle management and explanation to those affected.
• Generative AI: users must be told they are interacting with AI and generated content must be marked.
• Foreign companies: those above certain revenue or user thresholds must appoint a local representative.
• Fines and grace period: the fines provided go up to 30 million won, and authorities announced a grace period of about one year before full enforcement.

China: rules by type of service

China has no general AI law, but it does have mandatory, specific rules already in force: recommendation algorithms (2022), deep synthesis or deepfakes (2022), measures for generative AI services (2023) and measures for labelling AI-generated content, applicable since September 1, 2025. If your service reaches users in China, you will usually need to meet registration, security assessment and labelling requirements. Always check with local counsel.

Japan: promote, do not punish

The AI Promotion Act was passed in May 2025 and promulgated on June 4, 2025. The AI Strategy Headquarters and the basic plan provisions took effect on September 1, 2025, and the first AI Basic Plan was adopted on December 23, 2025. The law contains no fines or prohibitions: it asks companies to make an effort to use AI responsibly. The practical references are METI’s AI Guidelines for Business, which are non-binding. That does not exempt you from other laws (data protection, intellectual property, product safety), which can penalise.

Singapore, India and Australia: no AI-specific law

• Singapore: voluntary frameworks and sector guidance, for example in financial services.
• India: no AI law. The personal data protection law (DPDP) and its rules apply; they were notified in November 2025 with obligations phased in until May 2027, alongside voluntary government guidance.
• Australia: no specific law; a voluntary guide to AI adoption (October 2025) plus privacy law and sector regulators’ obligations.

What to do if your company operates in the region

1. Map where your users are. What matters is the user’s country, not where your company is.
2. Classify each system. If it decides about people (employment, credit, health), assume it will be “high-impact” in Korea and high-risk in the EU.
3. Build a common base: AI system inventory, risk assessment, technical documentation, AI-use notice and labelling of generated content. It serves the AI Act and most countries.
4. Watch the dates: Korea’s grace period and the phases of India’s data law are the nearest deadlines.

Frequently asked questions

Does the Korean law affect me if my company is European?

It can if you offer AI services to users in Korea and exceed the thresholds; you would then need to appoint a local representative. Confirm the current thresholds with local counsel.

Does Japan require anything?

There are no enforceable obligations in the AI law itself, but other rules that apply to AI, such as data protection or intellectual property, do bind.

Does complying with the AI Act help in Asia?

It gives you a good base (inventory, risk, documentation, transparency), but each country adds its own requirements, such as the local representative in Korea or registration in China.

Informational content, not legal advice. AI regulation changes quickly: always check the current official text before deciding. Information reviewed as of October 14, 2026.