The AI/Copyright rifts in SE Asia

There are three main areas for the creative and tech industry AI battles in SE Asia. All the major SE Asian markets are seeing frequent discussions, controversies and plans.

There are many examples of the issues in real life. In Vietnam in September, filmmakers of the “Mai No Mot Loi Tam Biet” (Chrysalis) film used AI and CGI to replace a female actress, who had been caught in scandal. Thailand's DIP has mediated disputes between artists and content creators whose songs, lyrics, or voice likenesses were adapted or cloned using AI tools. in Indonesia, disputes and grievances have arisen where AI tools have been used to replicate the visual identities, commercial logos, or distinct artistic styles of local brands as well as politicians and celebrities without consent. LLM training and whether to put in place a TDM exception to attract data centers are increasingly common in the region.

This summary sets out the state of play for AI and copyright in the major markets:

• Singapore: Singapore was the first with an explicit, statutory Text and Data Mining (TDM) exception under Section 244 of its Copyright Act. It allows the copying of copyrighted works for computational data analysis—including training LLM models—so long as the user has lawful access to the underlying work. Infringement occurs if an AI output reproduces a substantial part of an original copyrighted work. Guidelines from the Intellectual Property Office of Singapore (IPOS), indicate users and developers can be held liable depending on who directed or caused the generation of the infringing material. Copyright requires human intellectual creation. Fully AI-generated works without human intervention lack copyright protection. Works made with AI assistance may qualify if human creative contribution is sufficient.

  
  

• Vietnam:  Vietnam adopted an explicit exception allowing TDM and AI model training on legally published works. There are exceptions for bypassing paywalls/technical protection measures and requires that processing does not unreasonably prejudice the normal exploitation of the original work. Outputs that merely replicate learned artistic styles or abstract concepts are generally non-infringing. However, outputs bearing high structural or expressive similarity (e.g., 80–90% similar details or composition) to an existing work are treated as unauthorized derivative works. Purely AI-generated content does not qualify for copyright ownership; protection is strictly reserved for human authors.

  
  

• Indonesia: Proposed legislative reforms distinguish between non-commercial and commercial TDM. While non-commercial academic research is permitted, commercial AI training requires licensing arrangements, alongside proposed transparency rules requiring summaries of training datasets. Proposed rules address restrictions on the direct imitation of expressive styles and outputs that substantially replicate copyrighted content. Under proposed regulations, fully automated AI outputs are excluded from protection. AI-assisted works qualify for copyright only if significant human creative input is established.

  
  

• Malaysia: There is no explicit statutory TDM exception under the Copyright Act 1987. AI training currently relies on standard fair dealing interpretations or explicit licensing. Outputs mimicking styles or concepts are generally permitted, but outputs containing substantial passages or near-identical reproductions of protected works risk civil infringement claims. Requires human authorship under the Copyright Act 1987. Purely AI-generated outputs cannot be copyrighted, though human editing, prompt selection, and creative arrangement could establish protection.

• Thailand: The Copyright Act contains no express TDM exception for AI training. Mining protected works generally requires authorization unless it falls under strict fair use criteria. Unauthorized generation of outputs replicating or adapting protected works constitutes copyright infringement under Sections 6 and 27 of the Copyright Act. Copyright protection under the Copyright Act B.E. 2537 requires human origin. Pure AI outputs are unprotectable public domain material. Copyright protection under the Copyright Act B.E. 2537 requires human origin. Pure AI outputs are unprotectable public domain material.

  
  

• Philippines: No statutory TDM exception exists. Unauthorized copying of protected works into AI training datasets is generally analyzed under standard reproduction, adaptation, and fair use principles. If an output reproduces or publicly displays substantial elements of a protected work, it triggers claims for unauthorized reproduction or derivative creation.  Creation of copyrights requires human authorship so autonomous AI outputs cannot own or be granted copyright.

  
  

Countries are at different stages. Many are reviewing legislation now (Indonesia), others need to update their laws so initial discussions are under way (Malaysia, Thailand, Philippines). Singapore and Vietnam have relatively recently updated copyright laws. ASEAN's Working Group on Intellectual Property Cooperation (AWGIPC), the primary regional body handling IP alignment under the ASEAN IPR Action Plan—has actively brought AI into its agenda but no specific position exists yet.

The challenge is whether differing treatment will follow. Some countries trying to attract data centres and AI companies have TDM exceptions.  But there has been no litigation on LLM training in the region yet.  Infringing outputs is perhaps more of a concern but creators have not yet made claims.  There is an emerging pattern in the region, that humans must create copyright, but AI assisted creation might allow human authorship.  We can expect a lot more discussion, and law changes in time.