Thailand's Emerging Text and Data Mining Exception: Between Copyright Orthodoxy and AI Competitiveness
Editor’s note: We are delighted to feature a guest post by Dr. Saliltorn Thongmeensuk, who completed her LL.M. and LL.D. at Nagoya University under the supervision of Professor Suzuki. She now sits on Thailand's copyright reform working group, and she shares her first-hand perspective on the proposed text and data mining exception now under deliberation.
Guest Post by Dr. Saliltorn Thongmeensuk, Senior Research Fellow, Thailand Development Research Institute (TDRI); Member of the Thailand’s Working Group on Developing Recommendations for the Use of Copyrighted Works in Responsible Artificial Intelligence Development
One of the more consequential yet relatively underreported developments in Thailand's emerging AI governance framework is theongoing discussion surrounding a copyright exception for Text and Data Mining (TDM). While much of the public debate on AI regulation has focused on risk classification, AI sandboxes, and governance obligations, the TDM discussion arguably raises a more fundamental question: whether Thailand's copyright system is capable of supporting the data-intensive innovation model upon which modern artificial intelligence depends.
The issue emerged prominently during the development of the Draft Principles of the Artificial Intelligence Innovation Act being spearheaded by the Ministry of Digital Economy and Society (MDES) and the Electronic Transactions Development Agency (ETDA). As part of the drafting process, policymakers convened a dedicated focus group involving copyright scholars, legal practitioners, academics, and AI stakeholders to consider whether Thailand should introduce a specific copyright exception permitting the use of protected works for AI training and other forms of computational analysis. The inclusion of TDM within the draft principles signals an important shift in Thai policymaking. Rather than treating copyright and AI as separate policy domains, regulators increasingly recognize that copyright law may become one of the most significant determinants of Thailand's future AI ecosystem.
At first glance, the policy challenge appears straightforward. Generative AI systems depend on enormous quantities of data. Large language models, image generators, recommendation systems, and other machine learning applications derive their capabilities from analyzing vast corpora of text, images, audio, video, and other forms of information. In practical terms, obtaining individual licenses for every copyrighted work involved in AI training is often impossible. Modern AI development frequently involves billions of data points collected from millions of sources across multiple jurisdictions. The transaction costs alone can render comprehensive licensing commercially infeasible. Yet the process of AI training often requires reproducing copyrighted material, potentially implicating one of the core exclusive rights protected under copyright law. This creates a tension between the traditional copyright framework and the realities of contemporary AI development.
The challenge is not unique to Thailand. Indeed, it has become one of the defining copyright debates of the AI era. What makes the Thai discussion particularly interesting is that it takes place against a backdrop of long-standing uncertainty regarding the scope of copyright exceptions under Thai law.
Unlike the United States, Thailand does not explicitly recognize a broad, open-ended fair use doctrine. Section 32 of the Copyright Act B.E. 2537 establishes general principles requiring that an exception must not conflict with the normal exploitation of the work and must not unreasonably prejudice the legitimate interests of the copyright owner. However, Thai courts have historically disagreed on whether Section 32 paragraph 1 operates as an independent exception or merely establishes conditions governing the specific exceptions enumerated elsewhere in the Act. This debate has persisted for decades and remains unresolved.
The prevailing Supreme Court position is generally regarded as restrictive. In Supreme Court Judgment No. 1908/2546, the Court held that a defendant seeking to invoke a copyright exception must satisfy three cumulative requirements. First, the use must fall within one of the specific exceptions explicitly recognized by the Act. Second, the use must not conflict with the normal exploitation of the copyrighted work. Third, the use must not unreasonably prejudice the legitimate interests of the copyright owner. This interpretation effectively prevents Section 32 paragraph 1 from functioning as a general fair use clause comparable to that found in U.S. law. As a result, AI developers seeking to rely on a broad interpretation of fair use face significant legal uncertainty.
Recent developments, however, suggest that the picture may not be entirely settled. In Judgment No. 118/2563 (not yet available in digital form), the Central Intellectual Property and International Trade Court demonstrated a willingness to recognize uses extending beyond the explicit statutory exceptions where the use served a highly specialized purpose, had no meaningful impact on the market value of the copyrighted work, and did not interfere with the copyright owner's normal exploitation of the work. Although the judgment concerned plagiarism detection rather than AI training, the reasoning is noteworthy because it signals a more functional and economically oriented approach to copyright exceptions. Such reasoning bears similarities to arguments increasingly advanced in support of TDM activities.
Nevertheless, relying on judicial evolution alone may not be sufficient. AI developers, investors, and research institutions generally seek legal certainty rather than litigation-driven clarification. This explains why policymakers are increasingly examining whether Thailand should follow the growing number of jurisdictions that have adopted dedicated TDM exceptions.
Internationally, four broad approaches have emerged.
Japan represents the most permissive model. Following amendments to the Copyright Act that took effect in 2019, Article 30-4 permits the use of copyrighted works for information analysis purposes, including machine learning and AI training, regardless of whether the activity is commercial or non-commercial. The Japanese approach is built around the concept of "non-enjoyment use." However, the interpretation of this proviso, particularly its application to large-scale generative AI training, remains controversial in Japanese scholarship and policy discussions.
Because AI systems analyze works for computational purposes rather than human appreciation, Japanese lawmakers concluded that the traditional rationale for copyright protection is significantly weaker in this context. For AI developers, this creates an exceptionally favorable legal environment. For copyright owners, however, it raises concerns regarding uncompensated exploitation and the erosion of licensing markets.
The European Union adopted a more cautious approach through Articles 3 and 4 of the Copyright in the Digital Single Market Directive. Article 3 establishes a mandatory TDM exception for research organizations and cultural heritage institutions. Article 4 extends TDM privileges to all users, including commercial actors, but allows rightsholders to reserve their rights through machine-readable opt-out mechanisms. In theory, the framework seeks to balance innovation and copyright protection. In practice, however, implementation has proven considerably more complex. AI developers must determine whether content has been reserved, while rightsholders face challenges ensuring that reservations are effectively communicated. The subsequent AI Act has further complicated matters by introducing transparency obligations relating to copyrighted training materials. Together, these measures have created one of the world's most sophisticated but also most administratively burdensome TDM regimes.
The United Kingdom occupies an intermediate position. Since 2014, Section 29A of the Copyright, Designs and Patents Act has permitted TDM activities for non-commercial research. The exception applies to both public and private research entities and cannot be overridden by contract. However, it does not extend to commercial AI development. Attempts by the UK government to introduce a broader exception covering commercial uses ultimately encountered strong resistance from creative industries and were abandoned. The UK experience illustrates the political challenges associated with copyright reform in the AI era. While policymakers may view broader exceptions as necessary to support innovation, creative sectors often regard such proposals as threats to existing licensing markets.
The United States, by contrast, continues to rely primarily on judicial interpretation through the fair use doctrine. Landmark decisions such as Authors Guild v. Google recognized large-scale copying for search and analytical purposes as transformative use. Many AI developers argue that machine learning training should be treated similarly because the works are used to identify patterns rather than to substitute for the original expression. Yet the absence of a dedicated statutory framework means that legal certainty remains elusive. Ongoing litigation involving major AI developers demonstrates that the boundaries of fair use in the AI context remain contested.
The question confronting Thailand is therefore not whether a TDM exception is possible. Rather, it is which model Thailand should adopt.
A Japanese-style exception would likely provide the strongest boost to AI innovation. It would significantly reduce legal uncertainty, lower transaction costs, and improve Thailand's attractiveness as a destination for AI investment and research. Such a framework would align with broader government objectives aimed at promoting AI adoption, expanding digital infrastructure, and developing domestic AI capabilities.
Yet adopting the Japanese model would not be politically costless. Thailand's creative industries, publishers, media organizations, and collective management bodies would almost certainly raise concerns regarding uncompensated use of copyrighted works. Some may argue that AI companies derive substantial commercial value from copyrighted content and should therefore contribute financially through licensing arrangements.
An EU-style model may appear to offer a compromise. By preserving opt-out rights and licensing opportunities, policymakers can claim to protect both innovation and creator interests. However, the European experience demonstrates that compromise often comes at the expense of simplicity. Implementing opt-out systems requires technical infrastructure, awareness among rightsholders, and mechanisms for verifying compliance. These requirements may prove challenging in a jurisdiction where digital rights management systems remain comparatively underdeveloped.
From a policy perspective, the most important consideration may ultimately be Thailand's strategic economic objectives. Countries that are major exporters of copyrighted content often have stronger incentives to prioritize licensing opportunities. Countries seeking to expand domestic AI capabilities may place greater emphasis on reducing barriers to innovation. Thailand occupies an unusual position because it has interests on both sides of the equation. It seeks to promote creative industries while simultaneously aspiring to become a regional digital and AI hub.
This balancing exercise is further constrained by international copyright obligations. Any TDM exception must remain consistent with the Berne Convention and the TRIPS Agreement, particularly the three-step test requiring that exceptions be confined to certain special cases, not conflict with the normal exploitation of the work, and not unreasonably prejudice the legitimate interests of rightsholders. Much of the global debate over AI and copyright increasingly revolves around differing interpretations of these principles.
Ultimately, the significance of Thailand's TDM debate extends beyond copyright law. The issue sits at the intersection of innovation policy, industrial strategy, digital competitiveness, and international trade. Whether Thailand adopts a broad exception, a conditional exception, or no exception at all will send an important signal regarding the country's approach to AI development.
In that sense, the emerging discussion is not merely about whether machines should be allowed to read copyrighted works. It is about how Thailand intends to position itself in an increasingly data-driven global economy. The answer may shape the country's AI ecosystem for years to come. At the time of writing, no final policy decision has been reached, and there is no publicly announced timetable for legislative amendment. Nevertheless, the establishment of the working group and its ongoing comparative assessment indicate that Thailand is actively exploring possible reforms rather than maintaining the status quo. The eventual policy choice—whether to adopt a dedicated TDM exception, rely on existing copyright doctrines, or pursue a licensing-based solution—will play an important role in shaping Thailand's AI ecosystem in the coming years.