印刷する

[Vietnam Legal Update] "Key Highlights of the Decree Implementing the Law on Artificial Intelligence"

Key Highlights of the Decree Implementing the Law on Artificial Intelligence

May 28, 2026

1. Introduction

On 30 April 2026, the Vietnamese Government issued Decree No. 142/2026/ND-CP providing detailed regulations and measures for the implementation of the Law on Artificial Intelligence ( “Decree 142”). This is the first decree implementing the Law on Artificial Intelligence No. 134/2025/QH15, which came into effect on 1 March 2026 (“AI Law”). Decree 142 took effect on 1 May 2026, and several important highlights are summarized below.

2. Scope of regulation

Decree 142 has a broad scope of regulation as prescribed in Article 2. It regulates all entities involved in Artificial Intelligence (“AI”) systems, ranging from providers, developers, and deployers to users and those affected by AI systems. Notably, Decree 142 also applies to foreign organisations and individuals engaging in AI-related activities in Vietnam. This means that foreign individuals or technology companies deploying AI products or services in the Vietnamese market may also fall within the scope of regulation, regardless of whether their principal offices are located inside or outside Vietnam. However, the scope of the phrase "engaging in AI-related activities in Vietnam" remains ambiguous. For example, it is not expressly specified whether Decree 142 extends to situations where overseas providers directly supply AI-related services to customers in Vietnam or where AI systems are deployed within Vietnamese subsidiaries under the operation of their overseas parent companies.

3. Establishment of the Artificial Intelligence Single-window Portal and the National Database on Artificial Intelligence Systems

One of the notable highlights of Decree 142 is the first-ever establishment of two national-level AI governance infrastructures, managed and operated by the Ministry of Science and Technology (“MoST”), comprising:

- Artificial Intelligence Single-window Portal (AI Single-window Portal): Serves as a centralised platform for processing administrative procedures and obligations related to AI systems. Its key functions include the following:

・Receiving notifications of risk classification results and conformity assessment results;

・Issuing identification codes for AI systems;

・Receiving incident reports and periodic reports;

・Supporting automated risk classification of AI systems;

・Publicly disclosing AI-related information including conformity assessment results and violation handling outcomes;

・Receiving requests for guidance; and

・Monitoring the use of AI systems by state agencies

However, as of the date of this article, the portal has not been fully operational. Accordingly, enterprises should closely monitor announcements regarding its launch date through official press releases. In the meantime, enterprises should also designate personnel to familiarize themselves with matters such as AI system identification codes, electronic confirmation procedures, and related matters.

- National Database on AI Systems: Operated under the management of the MoST.

4. Mandatory Classification of AI Systems into 3 Risk Levels

AI systems are classified into three risk levels: high-risk, medium-risk, and low-risk (Article 9.1 of the AI Law). Providers are required to complete an assessment to determine which of these levels an AI system falls into prior to placing the system into operation (Article 5 of Decree 142).

In addition to the provider’s classification obligation, the deployer must cooperate with the provider to review and reclassify the AI system where deployment, modification, integration, or changes in function or intended purpose give rise to new or higher risks compared to the originally classified system.

The classification criteria for each risk level are as follows:

- High-risk AI systems: As of the date of this article, the list of high-risk AI systems has not yet been published. Since the list is expected to be issued in Q3 2026, businesses should closely monitor future official announcements.

- Medium-risk AI systems: Decree 142 expressly provides that an AI system will be classified as a medium-risk AI system if it satisfies both of the following conditions: (i) it does not fall within the list of high-risk AI systems; and (ii) it is capable of causing confusion, influencing, or manipulating users because they do not recognize that the interacting subject is an AI system or the content is generated by an AI system. It should be noted that the second condition – “capable of causing confusion, influence, or manipulation” – is broad and not clearly defined. Accordingly, a case-by-case assessment may be necessary to determine whether particular characteristics are present, including: simulation or impersonation functions; direct interaction with users; or direct provision of services or content to the public. Article 9.3 of Decree 142 also sets out specific exclusions.

- Low-risk AI system: AI systems that do not fall within either the medium-risk or high-risk categories.

5. Preparation of Risk Classification Dossiers and Conformity Assessment

Providers are required to prepare a risk classification dossier for high-risk and medium-risk AI systems prior to placing such systems into operation. The components of the dossier are prescribed in Article 12.2 of Decree 142.

For AI systems classified as high-risk, providers must conduct a conformity assessment before placing the system into operation. This is a mandatory obligation under Article 13 of the AI Law. Furthermore, throughout the operational period, if the system undergoes significant changes to an extent that may affect the initial assessment results, the provider must conduct a re-assessment of conformity.

6. Labelling Requirements for AI-Generated Content

Pursuant to Article 18 of Decree 142, deployers are obliged to affix clearly recognisable labels to audio, images, or videos generated or edited using AI systems in the following cases: (i) where such content simulates or impersonates the appearance or voice of a real person; and (ii) where such content recreates real events, for the purpose of distinguishing AI-related content from authentic content, unless otherwise provided by law.

However, Decree 142 also provides exemptions from the labelling requirement in certain cases, such as technical editing, text processing, internal use, or research and testing. Details of these exemptions are set out in Article 18.4 of the Decree.

7. Obligation to Report Serious AI System Incidents within 72 Hours

Article 19 of Decree 142 establishes a framework for reporting and handling serious incidents involving AI systems. Accordingly, in the event of a serious incident, the provider, or the deployer (where the provider cannot be contacted), shall submit a preliminary incident report through the One-Stop Electronic Portal on AI. The reporting deadlines are (i) within 72 hours for urgent and uncontrolled incidents; and (ii) within 5 working days for other serious incidents.

8. AI Sandbox

Decree 142 contains a dedicated chapter governing the controlled testing of AI systems, covering principles, classification of levels, competent authorities, approval procedures for participation, as well as conditions and documentation required for participation in the testing.

9. Recommendations for enterprises

Enterprises should implement internal action plans, including conducting a comprehensive review of all AI systems currently in use or to be deployed; carrying out risk classification and preparing risk classification dossiers; reviewing obligations relating to technical labelling of AI-generated content; and establishing incident response procedures for AI systems.

In addition, since the official launch date of the AI Single-window Portal has not yet been determined, enterprises should continue to monitor official announcements issued by the Vietnamese government and closely follow further developments in AI-related legislation and regulatory trends.

Date: 18 May 2026