Vietnam now has a binding legal framework for artificial intelligence. For foreign-invested enterprises, the new rules are relevant not only to companies developing AI, but also to businesses using AI tools in their day-to-day operations. The new regime is not limited to compliance obligations, but also introduces a controlled testing or sandbox regime.
Vietnam’s Law on Artificial Intelligence No. 134/2025/QH15 was passed on 10 December 2025 and took effect on 1 March 2026. It was followed by Decree 142/2026/ND-CP, effective from 1 May 2026, which provides the detailed implementing rules.
For most foreign-invested enterprises (“FIEs”), the practical question is not whether they develop AI themselves. Many already use AI in recruitment, customer service, fraud detection, credit assessment, marketing or internal group systems. These uses may be enough to bring the Vietnamese entity within the scope of the new rules.
This article looks at the main issues FIEs should consider when reviewing their use of AI in Vietnam.
The Law distinguishes between four roles: developer, provider, deployer and user. Importantly, these roles are determined for each AI system rather than for the company as a whole. The same company may therefore have different roles for different systems.
A developer designs, builds, trains, tests or fine-tunes an AI model. A provider puts an AI system on the market or into use under its own name or brand. A deployer uses an AI system under its control as part of its professional or business activities, while a user is the person who directly interacts with the system or uses its output.
For many FIEs, the distinction is relatively straightforward. If a Vietnamese subsidiary uses an AI tool developed or licensed by its parent company, the parent will generally be the developer and/or provider, while the Vietnamese subsidiary will be the deployer. The same applies where the company subscribes to a third-party AI or SaaS product and uses it internally.
The position changes where the Vietnamese entity incorporates AI into a product or service offered to customers under its own name. For example, an FIE operating an AI-enabled chatbot as part of its own banking, e-commerce or software service may itself qualify as a provider, even if the underlying AI model comes from its parent company or an external vendor.
This distinction matters because providers carry substantially more compliance obligations. These include classification and notification of certain AI systems and, for high-risk systems, risk management, technical documentation, human oversight and conformity assessment.
Deployer obligations are more operational. The company must use the system within its intended purpose and risk level, maintain appropriate data safeguards and human intervention, comply with applicable labelling requirements, record and report incidents and cooperate with the provider where inspection or remediation is required.
As a first step, an FIE should therefore identify the AI systems it uses in Vietnam and determine its role in relation to each one.
The new framework divides AI systems into high-, medium- and low-risk categories.
High-risk systems include those appearing on the list issued by the Prime Minister as well as systems meeting the criteria set out in Decree 142. Relevant factors include the potential for serious harm to life, health, property or individual rights, use in essential or public-interest sectors, the number of users and connections to critical infrastructure.
Medium-risk systems generally involve situations where users may be confused, influenced or manipulated because they do not realize that they are interacting with AI. Customer-facing chatbots and certain AI-generated content are examples.
There are, however, exceptions. Under Article 9 of Decree 142, a system may fall outside the medium-risk category where, for example, it merely modifies the form of existing content without creating new content, its AI nature is already obvious from the context, it does not directly interact with the public, or it only processes internal data without directly affecting external users.
Systems that do not fall into either category are treated as low risk.
The provider is responsible for classifying the system before deployment. Medium- and high-risk systems must also be notified to the Ministry of Science and Technology through the national one-stop AI portal.
For an FIE using a system supplied by its parent company or an external vendor, this means that classification should not simply be assumed. The Vietnamese entity should understand how the provider has classified the system and whether the relevant notification requirements have been completed.
High-risk AI systems require a conformity assessment before they can be used. Responsibility for that assessment sits with the provider: systems subject to mandatory certification require third-party certification, while other high-risk systems may be self-assessed.
For an FIE acting as a deployer, however, one provision deserves particular attention:
Where a high-risk AI system is operated in accordance with the applicable requirements but nevertheless causes damage, the deployer is required to compensate the injured party first. It may subsequently seek reimbursement from the provider, developer or other responsible parties. The deployer is exempt only where the damage results from the injured party’s intentional fault or force majeure.
This allocation of liability makes the contractual arrangements behind the AI system important. Where an FIE relies on technology supplied by a vendor or another group company, the agreement should address reimbursement and indemnification, cooperation in the event of an incident, access to technical documentation and evidence that the provider has completed the required classification and conformity assessment.
This is particularly relevant for businesses in finance, healthcare, education and other essential sectors, as well as companies using AI to make decisions affecting individuals, such as recruitment or credit decisions.
Another point for multinational groups is the requirement applicable to foreign providers.
Under Article 14.6 of the Law, a foreign provider whose high-risk AI system is used in Vietnam must maintain a lawful contact point in Vietnam. Where the system is subject to mandatory certification before use, the foreign provider must have a commercial presence or an authorized representative in Vietnam.
For multinational groups rolling out the same AI tools across several jurisdictions, this requirement should be considered when deciding how the system will be supplied to the Vietnamese subsidiary and which entity will assume the relevant compliance role locally.
The Law and Decree also introduce transparency requirements that may affect ordinary business use of generative AI.
Where an AI system interacts directly with individuals, users must be made aware that they are interacting with AI.
AI-generated audio, images and video must contain machine-readable marking, such as metadata, a watermark or a digital signature. Text is generally exempt unless another law requires otherwise.
There are additional requirements for content that simulates the appearance or voice of a real person or recreates real events. Such content must carry a visible label before it is released to the public.
Certain limited uses are excluded, including purely technical quality improvements, spelling or grammar correction, internal content that is not published and content used in a controlled testing environment.
These requirements are therefore relevant not only to technology teams, but also to marketing, HR and communications teams using generative AI for campaigns, avatars, videos or voice-overs.
Decree 142 introduces specific deadlines for reporting serious AI incidents.
A preliminary report must be submitted within 72 hours after confirmation of an incident involving harm to life or health, disruption of public services, or a rights violation that cannot be contained. Other serious incidents must be reported within five working days.
The filing of an incident report does not constitute an admission of liability.
Rather than creating an entirely separate procedure, companies may wish to incorporate AI incidents into their existing incident-response and data-breach processes, with clear responsibility for identifying whether the AI-specific reporting requirements have been triggered.
The new regime is not limited to compliance obligations.
Decree 142 introduces a controlled testing or sandbox regime. Depending on the sensitivity of the proposed testing, approval may be granted at provincial, ministerial or Ministry of Public Security level.
Participants may receive financial support of up to 50% of eligible actual costs for certain services, including computing infrastructure, shared data, AI models, technical consultancy and safety testing.
The Law also establishes a National AI Development Fund and provides support for conformity-assessment costs for SMEs.
FIEs carrying out AI-related R&D in Vietnam should therefore consider not only their compliance obligations but also whether their projects may qualify for these support mechanisms.
AI systems already operating before 1 March 2026 generally have 12 months, until 1 March 2027, to comply with the new framework.
A longer transition period applies in finance, healthcare and education, where compliance is required by 1 September 2027.
Systems may continue operating during the transition period, although the authorities may suspend a system that is likely to cause serious harm. Where an existing system is subsequently added to the high-risk list, Decree 142 allows up to 12 months to complete the required conformity assessment.
For most FIEs, the immediate exercise does not need to be complicated. The starting point is to identify the AI tools currently used in Vietnam, including tools supplied by the parent company and third-party vendors, and determine the company’s role for each system.
Particular attention should then be given to systems used externally or in decisions affecting employees, customers or other individuals. Companies should confirm the relevant risk classification with their providers, check whether notification or conformity assessment is required, and review contracts where the Vietnamese entity may bear liability as deployer.
Marketing and communications practices should also be reviewed where generative AI is used to create public-facing content, while existing incident-response procedures should be updated to cover the new AI reporting deadlines.
For many Vietnamese subsidiaries, the main compliance issue will therefore not be developing a new AI governance framework from scratch. It will be understanding which AI systems are already being used, who is legally responsible for them, and whether the company’s existing contracts and internal procedures adequately address the new rules.
Fidinam supports foreign-invested enterprises in Vietnam with understanding and addressing the legal and compliance implications of their business activities.
Our team can help you assess how Vietnam’s new AI framework applies to your existing AI systems, clarify your role and related obligations, and review relevant contracts, compliance procedures and risk management measures.
If your business is using or deploying AI in Vietnam, we can help you identify the key areas to address and prepare for the new requirements.
Contact us via info@fidinam.com.vn or submit the form below.