Law on Artificial Intelligence in Vietnam: Only regulates outputs and usage behaviors of artificial intelligence, does not hinder innovation

Risk-based approach to AI management: Harmonizing safety and development

Before proceeding to vote, the National Assembly heard Minister of Science and Technology (S&T) Nguyen Manh Hung, authorized by the Prime Minister, report to the National Assembly on the reception, explanation, and revision of the draft AI Law.

The Minister stated that the draft law is 20 pages long with 35 Articles. The content of the draft law's provisions has institutionalized the Party's new policy guidelines, meeting the requirements and objectives of law-making.

The Minister affirmed that AI is a new field that has recently seen very rapid development, impacting the economy, culture, society, security, and national governance methods at a deep level. Vietnam's early promulgation of the AI Law is necessary to create a legal framework for sustainable development, ensuring proactive opportunity seizure as well as risk management.

During the drafting process, the Government carefully studied international experiences, specifically: Europe manages strictly, focusing on safety; Japan mainly focuses on development; South Korea has both safety and development, but the safety portion is quite light.

"Vietnam chose a selective combined approach, ensuring harmony between both safety and development. Safety is higher than South Korea's basic level but lighter and involves fewer procedures than Europe's. Development is as strong as Japan's. This is a model suitable for Vietnam's conditions," Minister Nguyen Manh Hung affirmed.

Explaining a number of major contents, Minister Nguyen Manh Hung stated: Regarding the structure of the law, the draft is a concise framework law stipulating principles and the framework for state management of AI. Technical contents will be stipulated in decrees. This is an approach suitable for a technology field with a rapid pace of development.

Regarding AI models, the Law does not regulate models, because models are the internal innovation of enterprises. The Law only regulates outputs and AI usage behaviors as well as risks caused by AI to society. This is an international practice ensuring that innovation is not hindered.

Regarding management according to risk levels, the draft Law has reduced the levels from four to three: Low, Medium, and High risks, accompanied by an article on prohibited acts to ensure clear and transparent management stratification.

Explaining the inspection of high-risk AI systems, Minister Nguyen Manh Hung affirmed that the draft Law stipulates minimum obligations before deployment; enterprises must prepare safety assessment dossiers themselves but do not have to ask for permission; dossiers will have to be presented when the management agency performs post-audits.

Regarding compensation for damages, the draft has removed fines based on revenue. The Law stipulates compensation for all damages and includes sanctions for suspending service provision. These are measures strong enough and consistent with Vietnamese law.

Regarding whether high-risk AI systems are a source of high danger? The AI system in the draft is software with risks arising from decision-making, not physical risks inherent to itself like motor vehicles or explosives. Therefore, the draft Law does not classify high-risk AI systems into the group of highly dangerous sources. However, to enhance the protection of citizens, the draft Law adds a strict damage compensation mechanism, stipulating liability to compensate for damages to related subjects even without fault.

Regarding regulations on promoting development and innovation in the AI field, the draft Law sets out policies to promote transfer development, including: Developing AI infrastructure, expanding access to shared data, supporting the testing of new models, incentives for AI enterprises, developing human resources, and supporting small and medium-sized enterprises in applying AI. These policies aim to form an autonomous AI ecosystem and improve national competitiveness.

After hearing the explanatory report, the National Assembly proceeded to vote and passed the AI Law with 429/434 participating delegates voting in favor, accounting for 90.70%.

Human-centric: Safety, transparency, and responsibility

The AI Law consists of 35 articles, comprehensively covering everything from development principles, State policies, risk classification-management mechanisms, transparency requirements, and incident handling to conformity assessments for high-risk AI systems.

The consistent focus of the Law is human-centricity, ensuring safety, transparency, and accountability, while creating conditions to promote innovation in both the public and private sectors.

The AI Law requires maintaining human control over all important decisions of AI systems, limiting bias and discrimination; and encouraging the development of green, sustainable, and inclusive AI.

The State prioritizes the development of data-computing infrastructure, expanding shared data, supporting the testing of new technologies, promoting the innovation ecosystem, and applying AI in state management and public service provision.

The AI Law also sets requirements for strict management of AI in essential fields such as healthcare and education; ensuring data safety, privacy rights, and suitability to specific characteristics. Scientific research applying AI must comply with ethics and scientific integrity.

To protect citizens, the law prescribes a list of prohibited acts such as using AI to forge, manipulate perception, collect/process data illegally, or falsify information that is required to be transparent. Besides that, the law establishes a single-window Electronic Portal on AI and a National Database on AI systems, helping to increase transparency and support testing registration, risk classification, and incident reporting.

The AI Law also mandates transparency with content generated by AI: Audio, images, and video must bear machine-readable identification signs; products likely to cause confusion must be labeled.

In the event of serious incidents, subjects must timely record, notify, and remedy; the management agency may request the suspension or recall of the system.

The Law defines national AI infrastructure as strategic infrastructure, developed according to a unified - open - safe model, encouraging public-private partnership and the participation of enterprises, institutes, schools, and social organizations.

The State holds the role of directing, coordinating, and ensuring infrastructure capacity to serve national AI development; encouraging enterprises, research institutes, universities, and social organizations to invest, build, and share infrastructure; enhancing public-private cooperation in developing AI infrastructure...

With the passing of the AI Law, the National Assembly adds a key institutional foundation for the national digital transformation process. The Law not only creates a legal corridor ensuring safe and sustainable development but also paves the way for enterprises to boldly innovate, test, and compete using new technologies. The AI Law takes effect from March 1, 2026, marking a new step forward in AI governance, whilst clearly demonstrating Vietnam's orientation: Strong development but must be safe, innovation but must be responsible, and technology must serve humans.

Source: https://mst.gov.vn/luat-tri-tue-nhan-tao-chi-dieu-chinh-dau-ra-va-hanh-vi-su-dung-tri-tue-nhan-tao-khong-can-tro-doi-moi-sang-tao-197251210180352652.htm