Who Owns AI-Generated Content in Vietnam?
Excerpt: Vietnamese law now has an express rule on AI-generated content in Vietnam — but it is a rule that hands the question to the Government rather than answering it. Here is what is settled today, what is not, and what a company should put in its contracts in the meantime.
Start with a decided case, not the statute
Judgment No. 72/2023/KDTM-PT of 5 July 2023 — Superior People’s Court in Ho Chi Minh City. A United States software owner held a registered copyright in an engineering program. A Ministry of Culture, Sports and Tourism inspection had already found that the Vietnamese defendant copied the program onto a machine without permission. The infringement was accepted. The owner claimed USD 69,590 — and recovered nothing.
The reason matters more than the outcome. Under Articles 204 and 205 of the Intellectual Property Law, damage has to be proved and quantified separately from the infringing act. The defendant said an employee had installed the software himself; no drawing or product had been made with it; and the owner produced no evidence of commercial exploitation or benefit. Proving the copying proved only the copying.
Anyone planning to enforce rights over a model, a dataset or machine-generated output in Vietnam should read that as the working rule: the act and the loss are two separate evidential jobs, and the licence price is not the measure of the loss.
The legal basis for AI-generated content in Vietnam
- Law No. 131/2025/QH15, in force 1 April 2026, amending the Intellectual Property Law — Article 6(5) (rights in objects created using an AI system), Article 7(5) (the training-data exception), Article 22(1) (software supplied as a service).
- Intellectual Property Law, consolidated as No. 11/VBHN-VPQH — Articles 204 and 205 on remedies and the measure of damage.
- Vietnam’s dedicated Law on Artificial Intelligence and its implementing instruments — labelling, risk classification, controlled testing.
- The EU AI Act, where a Vietnamese company places a system on the EU market or its output is used there.
What Article 6(5) actually says about AI-generated content in Vietnam
Article 6 of the Intellectual Property Law sets out how rights arise — by creation, by registration, or by use, depending on the object. Law No. 131/2025/QH15 added a fifth clause to it. That clause does not state who owns machine output. It states that the Government shall regulate how the rules in Article 6(1) to (4) apply where the object was created using an artificial intelligence system.
So the position is this. The legislature has recognised that AI-assisted creation does not fit the existing categories cleanly, and has reserved the answer to a decree that has not yet been issued. Until it is, AI-assisted output falls back on the ordinary Article 6 rules, with all the uncertainty that implies about whether a human authorship threshold is met.
For a business, that is not a reason to wait. It is a reason to decide the question by contract.
Training data: the one clear permission
Article 7(5) is the provision most often relied on, and it is drafted narrowly. Organisations and individuals may use texts and data on intellectual property objects that have been lawfully published and that the public is permitted to access, for scientific research, testing and training artificial intelligence systems — provided the use does not unreasonably prejudice the rights and legitimate interests of the author or owner.
Three limbs, each of which has to be satisfied:
- the material was lawfully published;
- the public is permitted to access it;
- the use does not unreasonably prejudice the rightholder.
Material scraped from behind a paywall, or in breach of a site’s terms of use, fails the first two limbs before the third is reached. Note also that the exception is drafted by purpose, not by actor: a commercial developer may rely on it as readily as a research institute, so long as the purpose is research, testing or training.
What to put in the contract while the decree is pending
Because the statutory answer on ownership is incomplete, most real exposure around AI-generated content in Vietnam is allocated by agreement. Four provisions decide the outcome of nearly every dispute we see:
- Ownership of output. Who owns what the system produces, and what happens to derivative work built on it.
- Training-data warranties. Whether the supplier warrants that its training material met the Article 7(5) conditions.
- Indemnity scope. Whether third-party IP claims arising from output are covered — and whether the liability cap swallows the indemnity.
- Service terms over statute. For software delivered as a service, the fallback position under Article 22(1) as amended is the service terms, so those terms have to be read as though they were the statute.
Evidence to preserve for an AI-generated content claim in Vietnam
Judgment No. 72/2023/KDTM-PT is the reminder. If the day ever comes when you have to enforce, the file that wins is not the registration certificate — it is the record of deployment and revenue. Keep the training-data provenance log, the model version history, the deployment records and the revenue attributable to the output. Those are what turn an established infringement into a recoverable figure.
Where AI-generated content in Vietnam sits in our wider practice
This note is part of the firm’s work on AI and technology law. The full treatment — the training-data exception, the EU AI Act’s reach into Vietnamese operations, the dedicated AI statute, and deployment checklists — is on the AI lawyers in Vietnam page. Where AI sits inside a transaction, see Vietnam intellectual property lawyers.
Decided technology and intellectual property cases, with the reasoning, are published on case.unilaw.vn.
The contractual allocation of AI risk sits inside the ordinary law of contract, which is set out by our Vietnam contract team.