AI and Copyright landscape in Japan and Singapore

In August 2025, I participated as a panelist in a panel discussion titled “AI 101 for the bar and bench” hosted by the Internal Bar Association in Singapore. Both Japan and Singapore’s copyright laws have "clean" exceptions for the development of AI models, and the AI industry voiced support for the positive economic effects brought about by such legal certainty and stability. Japan and Singapore's AI policies play an important role in the international AI community. However, challenges remain in the global deployment of AI development and services. This newsletter provides an overview of the discussion and presents key takeaways.

1. Outline of Panel Discussion

• The session addressed the intersection of AI and copyright, providing practitioners with the technical foundation to better understand how AI models work and navigate copyright disputes in the age of AI. By doing so, this panel aimed to demystify AI and create a dialogue between technology experts, legal practitioners and industry leaders, to enable meaningful discussions on policy, governance and innovation.
• The history behind using AI and its proprietary data for over 40 years was shared. Access to broad and diverse training data is critical for data-driven science to achieve cost-efficient business operation. It is also essential to decrease repetitive tasks with greater freedom for creators to focus on creative works.
• AI is a broad set of technology, and the LLM is just part of it. Considering its complexity, legal certainty around AI is particularly important. Explicit and clear copyright exemptions to train AI create much of this legal certainty.
• An overview of the copyright exemption for information analysis under the Singapore Copyright Act is explained. The Singapore Copyright Act 2021 has an exemption for computational data analysis that is equivalent to the text data mining exemption as well as the fair use exemption. To qualify as copying or communicating for computational data analysis, the purpose (e.g., the copy is made for computational data analysis, and it is not used for any other purpose), usage (e.g., the copy is not provided to any person other than for verifying the results of the computational data analysis carried out), and lawful access (e.g., it has lawful access to the material (i.e., the first copy) from which the copy is made) would be examined.
• An overview of the copyright exemption for data analysis and minor use under the Japanese Copyright Act is also provided.
  • In Japan, in light of rapid technological developments, even before generative AI emerged, the Japanese government revised the Copyright Act in 2018. Although it did not introduce fair use provisions, it introduced broad restrictions on copyright regarding data analysis and minor uses of copyrighted works. In addition, in response to recent developments in AI, the Agency for Cultural Affairs of Japan issued the “AI and Copyright Guidelines” in March of 2024. While they are also not legally binding, they nevertheless provide important practical certainty for business operators, practitioners, and the Japanese courts.
  • Regarding the AI model training stage, the legal issue is whether Article 30-4 of the Copyright Act applies to the collection of learning data and processing of training data.
• Article 30-4 provides a broad exemption for use of copyrighted works for the purpose of data analysis, and this data analysis exemption can apply even to commercial purposes, which is why Japan has sometimes been referred to as a “machine learning paradise” or “machine learning heaven.” According to the AI and Copyright Guidelines, whether data collection and learning processes are using copyrighted works only for non-enjoyment purposes, which means not using creative expressions in copyright works, and not diluting their markets, must be considered.
• With respect to the AI output generation stage, the key requirement of copyright infringement is similarity with and reliance on existing copyrighted works. That said, if output falls within the scope of the minor use exemption under Article 47-5 of the Copyright Act, then the output does not constitute copyright infringement. This may be the case if the amount of output is fairly limited. In addition, data analysis for the output to which the minor use exemption applies is also exempted from copyright infringement at the same time.
• Recently major Japanese news publishers filed lawsuits against Perplexity AI in August 2025. ◻ This is the first copyright lawsuit between a major news publisher and an AI developer in Japan, and it is expected to be a landmark case for AI and copyright in Japan. Since it appears that Perplexity AI does not have an office in Japan, the complaint is expected to be served on its U.S. headquarters. Therefore, it may be more than a year or more before the first ruling is made by the Tokyo District Court.
• The Japanese government’s pro-AI/innovation policy approach is also explained.
  • The Japanese IP Promotion Plan 2025 states that the Japanese AI system market is expected to be worth more than one thousand billion Japanese yen, which is equivalent to approximately 9 billion Singapore dollars. As background to this recent development, regarding Japan’s overall approach to AI regulation, the Japanese government has adopted an innovation-first approach. That is, historically, Japan did not have any hard laws that comprehensively or specifically addressed AI; rather, the Japanese government adopted a soft-law-based approach. Thereafter, the Japanese government finally enacted the Japanese version of the AI Act in May 2025, which will become fully effective by this September. However, until now, the Japanese government’s basic stance has remained unchanged.
  • First, during the process of drafting the AI Act, the Japanese government engaged extensively with various AI stakeholders, regardless of whether the stakeholders were domestic or foreign companies or business associations. This process helped ensure that the provisions of the AI Act were in line with the reality of the AI industry and its dynamic market and did not come as a surprise to stakeholders.
  • Second, the AI Act imposes no penalties or third-party conformity assessment requirements. Instead, the primary objectives of the AI Act are to confirm the principles that should be ensured for responsible AI use and to ensure the Japanese government’s financial support of AI research and development in Japan.
  • Third, the AI Act also grants the Japanese government the legal authority to require the reporting of serious incidents caused by AI and to request information concerning AI broadly. These powers may initially be exercised over advanced AI model developers. However, they are expected to be used to determine the types and methods of governmental support required for fostering AI innovation in Japan, rather than to create new regulations.

2. Takeaways for Future Outlook

• Not only have Japan and Singapore shared pro-AI/innovation-first policy stances, but they have also established “clean” data analysis exemptions for copyrights that are crucially important for machine learning in AI model development. These shared AI policy advantages within the APAC region and the ASEAN-Japan cooperation framework are expected to serve as a beneficial reference for other jurisdictions and generate spillover effects for them.
• Human-touch nature in product design is still important for generating goods or services and it is an exaggeration to say that AI will immediately replace humans. We need to continue considering how to effectively use AI and the regulatory framework. In reality, the law lags behind the technology; therefore, both private companies and governmental authorities are responsible for the development and improvement of the regulatory framework for AI.
• Fair use provisions and the other copyright exemptions are not always clear. Also, legal certainty may not necessarily deter investment in AI; rather, the other broader factors may affect and attract investments in AI infrastructure such as foreign direct investment regulations. AI policy must be constructed based on a comprehensive legal and policy framework.
• In-depth understanding of technology is also required. AI does not necessarily rely or depend on existing copyrighted works to generate output; rather, AI reconstructs and generates new original content based on parameters. Regulatory frameworks based on an insufficient understanding of AI technology risk becoming excessively precautionary and restrictive, potentially leading to irreversible damage to economic development opportunities.
• How we can build an internationally harmonized framework in the copyright field may be critical for global AI business. International forums such as WIPO may spearhead this type of discussion, but it is not certain. The choice of governing law or conflict-of-law system may reduce uncertainty to some extent, but the risk may remain that an individual country with a major market creates an unharmonized framework[Ed: please confirm, if necessary] that leads to “court shopping.” As a conflict-of-law matter, the text data mining exemption may not be considered a mandatory rule.