Vietnam's Artificial Intelligence Law, effective from 1 March 2026, requires foreign providers of high-risk systems supplied in Vietnam to maintain a local legal contact. Where a system requires mandatory conformity certification before use, the provider must have a commercial presence or authorised representative. The distinction matters to overseas software suppliers deciding how to serve the Vietnamese market.1
The law also provides transition periods for systems operating before it took effect: eighteen months for healthcare, education and finance, and twelve months for other fields. Such systems may continue operating during the transition unless authorities determine that serious harm risks justify suspension or termination. A pre-existing service and a new deployment therefore require different timing assessments.1
The government describes supervision proportionate to risk
At commencement, the Ministry of Science and Technology described a framework with low, medium and high risk categories. It said the purpose was to enable innovation while protecting people and national interests, with stricter transparency, safety and accountability requirements for higher-risk systems. That is the government's stated policy rationale, rather than evidence of how much compliance will cost individual suppliers.2
The ministry's explanation emphasised:
AI chỉ là công cụ hỗ trợ
AI is only a supporting tool.2
Its account places human responsibility alongside technical development, making governance part of the product's route to market.2
Implementation addresses information supplied by vendors
The government's register records that implementing Decree 142/2026 took effect on 1 May 2026. Businesses therefore have an implementing instrument to read alongside the statute, rather than only the framework announced at commencement.3
The decree requires risk-classification dossiers for medium- and high-risk systems. Where a provider builds on another company's platform, model or AI service, the required technical and data information is limited to what the provider can lawfully access and control. The dossier rules do not generally require disclosure of source code, model parameters or raw training data, subject to other legal provisions.4
They also allow suitable technical descriptions prepared under international standards if those documents satisfy the decree's information requirements. For a multinational supplier, that creates a way to use existing documentation, but does not establish automatic acceptance of every overseas compliance package.4
Duties continue after a system is supplied
Allen & Gledhill's March analysis highlights continuing duties for high-risk suppliers, including risk management, technical documentation, operational logs and information for inspections. Deployers also have responsibilities for safe operation, confidentiality and preventing unauthorised interference. The firm's account shows why responsibility does not rest solely with the company that originally developed the model.5
For medium-risk systems, it describes explanations to authorities covering intended use, functional operation, input-data types and risk controls. These are product-governance requirements relevant to contracts between foreign suppliers and local customers. The practical question for an intended deployment remains which risk classification and certification requirements apply to that system, and which party can supply the necessary evidence.5