On 15 October 2026, Vietnam's Cultural Industries and Copyright Office, under the Ministry of Culture, Sports and Tourism, held a national conference reviewing how the copyright provisions of the Intellectual Property Law worked from 2021 to 2026. According to the news report that prompted this analysis, participants concluded that the framework is struggling to keep pace with AI and cross-border digital platforms, and floated a standalone Copyright Law separate from the IP Law. The idea is at an early stage. Proposed topics include authorship of AI-assisted works, how much human creative input is needed for protection, and how works and data may be used for AI research and training.
The timing is notable. The amended IP Law (Law 131/2025/QH15) took effect on 1 April 2026, and Decree 134/2026/ND-CP followed on 9 April. A review of rules that are barely six months old deserves scrutiny, because the right question is not whether Vietnam should write a new statute but what a new statute would fix that the current package cannot.
The strongest case for a new law
The argument for reform is serious. The IP Law bundles copyright, industrial property and plant varieties into one instrument, so every copyright amendment must move through a vehicle designed for patents and trademarks. Copyright is also where AI pressure is most acute: generative systems are trained on creative works, and creators fear their output will be undercut by machine-made substitutes. A dedicated copyright statute could give authors, collecting societies and developers a single, readable text, and cross-border platforms a clearer counterparty. Regulators who say the framework lags the technology are not wrong that the pace of change is high.
What the current package already does
The existing rules are more specific than the phrase "struggling to keep pace" suggests. Per Baker McKenzie's analysis, Decree 134/2026/ND-CP, which amends Decree 17/2023/ND-CP, took effect on 9 April 2026 and added an Article 5a on AI-assisted works. Copyright or related rights arise only where a human makes a "substantial and decisive intellectual contribution" to the creative process. Tilleke & Gibbins adds two further conditions: the human must exercise effective control over the outcome and take responsibility for the content and its legality. Purely AI-generated output does not qualify.
On training, new Articles 37a to 37c set conditions for text and data mining. Baker McKenzie reports that Article 37a permits mining for scientific research, testing and AI training when the sources are lawfully published and lawfully accessed, technological protection measures are not circumvented, and the use neither conflicts with normal exploitation of the works nor substitutes for them in the market. Article 37b lets rightsholders reserve their rights through machine-readable signals or public declarations, though the reservation does not bite where Article 37a is fully met. Article 37c requires record-keeping and says royalties may apply when AI systems trained on such data are commercially exploited. The Can Tho provincial government's legal dissemination page confirms the statute itself permits use of lawfully published texts and data for research, testing and AI training, as long as authors' interests are not unreasonably harmed, and leaves detail to government rules.
So Vietnam already has an authorship test and a training framework with a rights-reservation mechanism. What it lacks is operating experience. Neither the 'substantial and decisive' threshold nor the 'normal exploitation' limit has been tested in a dispute that this article's sources report.
Where the real risk lies
The danger of an early rewrite is instability. Developers deciding whether to train or fine-tune models in Vietnam need to know which rules will govern a dataset assembled today. Rewriting the framework within a year invites the question of whether the 2026 conditions will survive. Startups and universities, which have the most to gain from lawful mining for research and the least capacity to track shifting compliance duties, would bear that uncertainty first.
The documentation burden is already notable. Baker McKenzie reports that creators must retain prompts, inputs and intermediate drafts and disclose AI use on request, and that miners must keep technical records and training datasets for authorities. These obligations are manageable for a large publisher and heavy for a student studio. A standalone law should not stack further duties on top; it should define the human-input threshold with examples, so that a designer who edits and curates AI output knows where the line falls.
The open-internet case matters too. The limits of copyright over data are a global fight. In the United States, the Ninth Circuit ruled on 16 September 2026 that stripping-of-attribution claims under DMCA Section 1202 cannot rest merely on AI output lacking copyright management information, as the EFF reported in a case over code used to train language models. The lesson for Vietnam is that liability theories should attach to actual copying, not to the mere fact of training or to the absence of metadata. Narrow, evidence-based liability protects creators from real substitution without chilling research.
What a good standalone law would do
- Keep the human-contribution test, and add worked examples across text, image, music and code, so the threshold is a standard people can apply rather than a discretion regulators exercise.
- Preserve the research and training exception and tie any royalty to demonstrated market substitution, not to the fact of training.
- Make rights reservation workable: a recognised machine-readable standard, so rightsholders can opt out cheaply and developers can comply automatically.
- Sequence the reform: collect case evidence from the 2026 rules, publish an impact assessment, and consult developers and creators before drafting.
A separate Copyright Law is a defensible long-term goal. But the useful reading of the 15 October conference is as a call to measure how Decree 134 performs before replacing it. If the evidence shows creators are being displaced, tighten the rules where the harm is. If it shows researchers are being deterred, widen the exception. Either way, Vietnam's advantage is that it has written down a testable rule early. It should learn from it rather than reopen it by reflex.