Asia tightens AI rules
Asian governments are accelerating AI governance—Vietnam has enacted a new risk‑based AI law that increases state oversight and mirrors parts of the EU approach while several countries are piloting controlled testing zones for emerging systems. At the same time, calls are growing for military guardrails: lawmakers are pushing proposed US legislation to ensure human oversight of AI in defense, signaling regulatory scrutiny is expanding beyond civilian use. ( )
Vietnam’s Law on Artificial Intelligence (No. 134/2025/QH15) was adopted by the National Assembly on December 10, 2025 with 429 of 434 deputies voting in favor and took effect on March 1, 2026. The statute codifies a three-tier risk classification for AI systems and grants most existing providers a grace period until March 1, 2027 (extended to September 1, 2027 for health, education and finance) to meet compliance requirements. The law replaces AI provisions in the 2025 Law on Digital Technology Industry, centralises state governance of “AI-related activities” while explicitly excluding activities solely for national defence, security and cryptography from its scope. Singapore’s Infocomm Media Development Authority published results from its Global AI Assurance Pilot and released a Testing Starter Kit after running technical tests of real‑world generative AI applications in 2025. Japan, South Korea and Singapore are explicitly using regulatory sandboxes or city-scale “testbed” programs to let firms run live AI experiments under limited exemptions, with Seoul’s “Testbed Seoul” allocating city space and grants to AI demonstrations in 2025. Analysts estimate a multi‑billion‑dollar compliance burden as Asian frameworks diverge in scope and technical obligations, raising costs for multinational AI vendors operating across the region. In Washington, Senator Elissa Slotkin introduced the AI Guardrails Act on March 17, 2026 to ban the Pentagon from firing autonomous lethal weapons without human authorization and from using AI to surveil Americans or to initiate nuclear launches, while FY2026 defense authorization text also contains multiple AI oversight provisions.
