Facing the urgent requirement to effectively manage new technologies and keep up with the world's development trends, Vietnam is gradually building and perfecting a separate legal framework for artificial intelligence. This is not only a necessary step to ensure safety and transparency in the application of artificial intelligence in multiple fields and industries, but also to create a stable legal corridor, promote innovation and support entities to participate in the digital technology ecosystem.
The 2nd Draft Law on Artificial Intelligence ("Draft") dated 21/11/2025[1] was submitted at the 10th Session of the 15th National Assembly, there have been many changes including 08 Chapters with 36 articles, refined compared to the 1st time. The content of this Draft inherits the regulations related to artificial intelligence in the Law on Digital Technology Industry 2025, revising in accordance with regulations focusing on relevant individuals and organizations and state management of artificial intelligence activities in Vietnam. Here are some typical points in the Draft Law on Artificial Intelligence.
Maintaining the risk classification of artificial intelligence systems
Classification of the level of risk of artificial intelligence systems is the premise and basis for determining the level of risks that can be caused by artificial intelligence in order to provide a legal framework to adjust in accordance with reality. In Article 7 of the Draft, the classification mechanism continues to maintain 04 gradually increasing risk levels with the system, including: (i) low risk; (ii) medium risk; (iii) high risk; (iv) unacceptable risks. This is considered a mandatory responsibility for suppliers before being circulated or put into use. In particular, for systems classified as high or medium risk, the classification results must be notified to the Ministry of Science and Technology through the one-stop web portal on artificial intelligence. In addition, the Draft encourages parties to develop low-risk artificial intelligence systems to disclose basic information on the one-stop web portal on artificial intelligence.
In some cases where the level of risk cannot be determined, the supplier may request the Ministry of Science and Technology to confirm the classification level on the basis of technical documents.
In addition, the purpose of this classification helps the management agency determine the level of fulfillment of obligations and legal responsibilities of individuals and organizations developing artificial intelligence systems to ensure early detection of risks and handling violations of deviations in the process of operating the system.
Legal obligations on transparency, labeling, and accountability of stakeholders to artificial intelligence systems
The responsibility for risk classification is considered the foundation in the risk management of artificial intelligence systems of state management agencies, the responsibility for transparency, cohesion and accountability continues to be regulated to implement, clarify and ensure safety. trust for users.
For labeling, the legal term "labeling" was first introduced in Clause 12, Article 3 of the Draft, creating a basis for the identification and management of information generated by artificial intelligence in many different contexts. This mechanism is a technical measure to notify users to recognize and avoid confusion of content or products created or edited to a significant extent by artificial intelligence.
However, not all intellectual systems are required to be labeled, but only the cases specified in Clause 3, Article 9 of the Draft are required to be labeled.[2]
Specifically, the system implementer needs to clearly inform users about the content generated by artificial intelligence, edited, processed, or provided when it is in cases where labeling is required. At the same time, the implementing party must label or disclose the attached information.
The supplier should be responsible for ensuring that the system is designed and operated in accordance with the previously identified risk level, accompanied by providing monitoring methods, technical guidance and necessary issues for the implementer to continue to fulfill the obligation of transparency. smooth and accountable.
Conditioning through a Controlled Testing Mechanism for Artificial Intelligence
Pursuant to Decision No. 127/QD-TTg dated January 26, 2021, the tasks of perfecting institutions and policies for the development of artificial intelligence are set such as: "Building a sandbox"; "Improving institutions and policies to attract foreign direct investment". To achieve that goal, in Article 22 of the Draft, a controlled testing mechanism is allowed to be implemented, which encourages the evaluation of artificial intelligence systems such as multi-agent artificial intelligence systems, automated artificial intelligence systems, etc ... to test the effectiveness and safety before commercialization.
Test results duly recognized by competent State agencies are considered as a basis for organizations and individuals to reduce the burden of procedures for conformity assessment, certification or approval to deploy the system. At the same time, these enterprises are also considered for exemption and reduction of some compliance obligations and priority access to financial support from the National Artificial Intelligence Development Fund.
This mechanism not only has many incentives for individuals and businesses that have been, are and are expected to operate in the field of artificial intelligence, but also serve as a basis and practical experience, helping to quickly improve standards and regulations with policies and legal regulations on artificial intelligence in the future. From there, promoting the State's cooperation with businesses and experts in the industry, developing an innovation ecosystem for the country.
Liability for damage caused by artificial intelligence systems
The legal gap in the responsibility of artificial intelligence is one of the most concerning issues for individuals and businesses operating in this field. Regulations on this issue will directly impact stakeholders in the value chain when artificial intelligence risks are still latent and arise at any time.
Along with that, from the perspective of the State management agency, the construction of a complete and appropriate legal corridor poses many challenges when determining legal liability from the damaging behavior of artificial intelligence. Because the consequences of this problem can be affected by many related subjects such as developers, manufacturers, users or the artificial intelligence system itself.
Currently, depending on the nature and severity of the consequences, individuals or businesses may be administratively sanctioned and/or prosecuted for criminal liability in the field of artificial intelligence. For administrative sanctions, the implementation is based on the provisions of specialized laws on artificial intelligence and the law on administrative handling. According to Article 30 of the Draft, enterprises may be fined up to 2% of their revenue in Vietnam and an additional fine of up to 2% of their global revenue in the preceding year if they continue to commit repeat offenses. However, in some cases where enterprises or individuals are newly established and have no revenue in the previous year, the sanctioning level may be up to 2,000,000,000 VND (two billion VND) for organizations and 1,000,000,000 VND (one billion VND) for individuals.
In addition, organizations and individuals are also responsible for compensating for damage caused by acts caused by artificial intelligence systems. Depending on the classification of the risk level of the system, the compensation liability will be different according to Clause 3, Article 30 of the Draft. For high-risk artificial intelligence systems, in case the system itself causes damage, the compensation will be determined due to the source of extreme danger according to Article 601 of the Civil Code 2015. On the contrary, damage caused by violations in the management and use of organizations and individuals shall be compensated by such organizations and individuals. For medium or low-risk systems, compensation shall be made according to the provisions of civil law. In case the artificial intelligence system is formed by many parties participating and deployed, the relevant parties may have to jointly compensate based on the level of management, control, and risk handling measures in practice.
The draft Law on Artificial Intelligence marks an important step in establishing a modern governance mechanism, in accordance with the characteristics of new technologies, as well as the trend of artificial intelligence development globally. The State, as a management agency, needs to "neutralize" between the two sides of new problems arising in society, between freedom of development and the risks brought. Obviously, this is not an easy "problem" for any country in the world, not only Vietnam.
Regulations on risk classification, transparency, labeling, controlled testing mechanisms and liability contribute to the formation of an adequate legal foundation, creating a safe and conducive environment for innovation. When finalized and promulgated, this law is expected to support businesses to develop sustainably, while ensuring the interests of users and the general interests of society.
Lawyer Nguyen Nhat Duong - Nguyen Viet Hung
HM&P Law Firm
Read more: Những điểm quan trọng trong Dự thảo Luật Trí tuệ nhân tạo
[1] https://duthaoonline.quochoi.vn/dt/luat-tri-tue-nhan-tao/251009091536864496, last accessed on 30/11/2025
[2] Clause 3, Article 9 of the Draft:
"3. Labeling is carried out in the following cases:
(a) Content generated or edited by an artificial intelligence system that is false, imitates real people, real events, and is likely to mislead viewers, listeners, or readers as real;
(b) Content generated by artificial intelligence for the purpose of communication, advertising, propaganda or public information;
(c) Other cases as prescribed by the Prime Minister."
