Collecting Vietnamese online data to train AI models is something many FDI tech teams do daily — mostly on the assumption that ‘public data is free to use’. Vietnamese law currently has no dedicated exception for AI training data mining, so that assumption needs re-testing against copyright, personal data and platform terms.
Three legal fences around AI training data
- The copyright fence: expression is protected, facts are not — online articles, photos and videos are works under Article 14 of the IP Law; bulk copying to build a training corpus is reproduction needing rights, while pure facts and ideas are unprotected.
- Vietnam’s exceptions are a closed list — Article 25 enumerates free-use cases (private study, reasonable quotation, teaching illustration…); Vietnam has no text-and-data-mining exception for commercial AI training — it cannot be imported by analogy from foreign law.
- The personal data fence — scraped data containing personal information (faces, identifiable posts) falls under the Personal Data Protection Law and Decree 356/2025/ND-CP, replacing Decree 13/2023/ND-CP from 1 January 2026: training use is processing needing a lawful basis.
- The platform contract fence — terms of news sites, social networks and marketplaces typically ban automated scraping; breaching them can mean blocked access, account termination and damages independently of copyright.
A practical compliance frame for AI data teams
- Prioritise clearly licensed sources: commercially licensed open data, self-generated data, customer data under contract, and verified public-domain corpora.
- Keep a provenance log for every dataset: source, licence, collection date, personal-data filtering — mandatory material when investors or enterprise clients audit the model.
- Build exclusion and removal mechanisms: an opt-out source list, a deletion process on rights-holder request, and data-protection impact assessments for sensitive datasets.
Good AI training starts with legally clean data, not just technically clean data. FDI AI builders in Vietnam should accept that the law offers no free pass for training data yet — and turn data governance into a selling point for enterprise clients who increasingly audit exactly this.
Frequently asked questions
Does scraping Vietnamese websites to train AI infringe copyright?
It can: copying works (articles, photos, videos) to build a training corpus is reproduction under the IP Law, and Article 25 has no exception for commercial AI training.
Which data is usable and which is not?
Safe: clearly commercially licensed data, verified public-domain works, pure facts, self-generated data. Risky: works still in copyright and data containing personal information.
Do we need the data owners’ permission?
For in-copyright works at commercial scale: a licence or contractual basis; for personal data: a lawful basis under Decree 356/2025/ND-CP.
This article has been quality-reviewed by the FLAW AI Legal Council
