Vietnam is the first country in Southeast Asia to have a dedicated law on artificial intelligence. The Law on Artificial Intelligence 134/2025/QH15 (passed 10/12/2025, effective 01/3/2026) has 35 articles, establishing a comprehensive framework for AI research, development and application. This article analyzes the key points and their impact on tech enterprises.
1. The philosophy of the law: AI serves humans, not replaces them
The law sets a foundational principle: “AI serves humans, not replaces them”. Accordingly, material decisions affecting the lawful rights and interests of individuals must be made under human supervision. This is the guiding direction for all compliance obligations businesses must build.
2. Three-tier risk classification
The law classifies AI systems into three risk tiers: high, medium and low. Obligations of AI developers and deployers increase with the risk tier. For high-risk AI in 6 sectors listed under Decision 33/2026/QD-TTg, businesses must fully perform compliance, notification and risk-classification obligations.
3. AI sandbox and transition period
The law provides an AI sandbox mechanism — a controlled testing environment for new AI systems before wide deployment. It also sets a transition period for AI systems already operating before 01/3/2026: businesses running AI systems must review and adjust them to meet the new requirements within the transition period.
4. Copyright in AI-assisted works
Decree 134/2026/ND-CP (issued 6/4/2026, effective 9/4/2026) regulates copyright in works involving AI. The core principle: humans are recognized as authors when their contribution is significant and decisive in the creative process. Businesses producing AI-assisted content should preserve evidence of human creative input to protect their rights.
5. Cybersecurity Law 2025: unlawful deepfakes prohibited
The Cybersecurity Law 116/2025/QH15 (effective 01/7/2026) prohibits unlawful deepfakes. Businesses using generative image, audio and video technologies need control procedures to avoid violating this rule.
6. Obligations of tech enterprises
Decree 142/2026/ND-CP (issued 30/4/2026, effective 01/5/2026) guides implementation of the AI Law. Note: this is an AI instrument — not to be confused with Decree 142/2017/ND-CP on maritime affairs (same number). Practical impact on tech enterprises:
- Classify their AI systems by the three risk tiers and determine whether they fall within the 6 high-risk sectors;
- Fulfil notification obligations for high-risk AI systems;
- Build human supervision procedures for material AI-assisted decisions;
- Review systems operating before 01/3/2026 within the transition period.
Frequently asked questions
When did the AI Law take effect?
Law 134/2025/QH15 took effect on 01/3/2026.
How is AI risk classified?
Three tiers: high, medium, low; high-risk AI in 6 sectors under Decision 33/2026/QD-TTg bears full compliance obligations.
What sector does Decree 142/2026 cover?
Implementation of the AI Law (not Decree 142/2017 on maritime affairs).
Can humans be recognized as authors of AI-assisted works?
Yes, when their contribution is significant and decisive in the creative process (Decree 134/2026/ND-CP).
Are deepfakes prohibited?
The Cybersecurity Law 116/2025/QH15 (effective 01/7/2026) prohibits unlawful deepfakes.
Sources
- Law on Artificial Intelligence 134/2025/QH15 (passed 10/12/2025, effective 01/3/2026) — 35 articles;
- Decree 142/2026/ND-CP (30/4/2026, effective 01/5/2026) — guiding the AI Law;
- Decision 33/2026/QD-TTg — list of high-risk AI (6 sectors);
- Decree 134/2026/ND-CP (6/4/2026, effective 9/4/2026) — copyright in AI-assisted works;
- Cybersecurity Law 116/2025/QH15 (effective 01/7/2026) — prohibiting unlawful deepfakes.
Conclusion
The AI Law 134/2025/QH15 makes Vietnam the first Southeast Asian country with a dedicated AI statute, built on the philosophy “AI serves humans, not replaces them” and a three-tier risk framework. Together with Decree 142/2026 guiding implementation, Decision 33/2026 on high-risk AI, Decree 134/2026 on AI copyright, and the Cybersecurity Law 116/2025 prohibiting unlawful deepfakes, tech enterprises face a new set of compliance obligations: risk classification, notification, human supervision, and review of existing systems within the transition period.
Does your tech business need an AI Law compliance assessment? Contact FLAT Law Firm for advice.
For reference only; not legal advice. This content was prepared by AI and will undergo quality review by the AI Legal Council of FLAT Law Firm before any publication consideration; it is not the legal opinion of a licensed human lawyer. Laws and regulations may change; verify the current legal texts before applying them.