The AI rules calendar for customer-facing AI in Asia-Pacific, 2025 to 2028

The calendar

Every date below comes from the regulator or from legal commentary on the official text. Several items are guidance, and the entry says so.

1. 4 Jun 2025Japan

   AI Promotion Act promulgated

   Passed by the Diet in May 2025. It sets up an AI strategy headquarters and a basic plan, and places cooperation duties on businesses without penalties.⁷

2. 10 Dec 2025Vietnam

   Law on Artificial Intelligence passed

   A single legal framework for developing and deploying AI, in force from 1 March 2026.⁸

3. 22 Jan 2026South Korea

   AI Basic Act and Enforcement Decree in force

   Includes transparency duties and a domestic representative requirement for foreign operators above set thresholds. The government has signalled a grace period of at least a year before fines.⁶

4. 22 Jan 2026Singapore

   Model AI Governance Framework for Agentic AI

   IMDA guidance launched at Davos, organised around four dimensions from bounding risk upfront to end-user responsibility. A living document.¹²

5. Jan 2026Singapore

   Starter Kit for Testing LLM-Based Applications, v1.0

   IMDA’s voluntary guide to pre-deployment testing of LLM applications for baseline safety and reliability.⁴

6. 2 Aug 2026European Union

   AI Act transparency duties (Article 50)

   People must be told when they are dealing with an AI system. Unchanged by the Digital Omnibus.⁹¹⁰

7. 7 Oct 2026Singapore

   MAS Guidelines on AI Risk Management issued

   Apply to all financial institutions and all forms of AI, with board oversight, an AI inventory, risk materiality assessment and controls across the life cycle.³

8. 14 Oct 2026Singapore

   Starter Kit for Testing Agentic Systems, v1.0

   IMDA’s practical reference for testing AI agents, launched with an AI tester accreditation programme.⁵¹¹

9. 2 Dec 2027European Union

   Annex III high-risk obligations

   Moved from August 2026 by the Digital Omnibus.⁹¹⁰

10. 7 Oct 2027Singapore

    MAS guidelines take effect

    Governance, oversight and core risk management expectations apply from this date.³

11. 7 Oct 2028Singapore

    MAS guidelines fully phased in

    The remaining sections, covering life cycle controls and capability, are expected by this date.³

What the Singapore dates mean for service teams

The MAS guidelines name agentic AI systems that operate autonomously and access tools, and MAS intends to consult the sector on further agentic AI guidance in 2027.³ A bank or insurer running a conversational agent should expect that agent in its AI inventory, with a materiality rating and controls in proportion. Institutions remain accountable for AI from third parties used in their services.³

The IMDA framework asks organisations to bound an agent’s autonomy and its access to tools and data, and to set checkpoints where a person must approve.¹ The testing kits give a method for checking both before launch.⁴⁵

Three different approaches in North Asia and Vietnam

Korea’s act carries obligations and a representative requirement for foreign operators, with enforcement eased during a grace period.⁶ Japan’s act promotes AI and relies on cooperation.⁷ Vietnam’s law sets a risk-based framework that took effect on 1 March 2026.⁸ A regional service operation needs a short note per market on which of these apply to its channels.