When AI Imitates, Replaces, and Steals: Rethinking Intellectual Property in Asia’s Emerging Creative Economies

A Juxtaposition of the Past

In the late fifteenth century, as the printed word began to eclipse the illuminated manuscript, a storm of intellectual unease swept across Europe. One storm of intellectual unease rumbled in the mind of Filippo. He admired books but despised printers. To Filippo, printers were not artisans but exceptional and foreign interlopers. He considered them to be lowly, typical, awfully commercial, beggar-thieves that had no care for language or learning, only a relentless hunger for money. Filippo believed they had filled the marketplace with their ugly tongue from press and rubrics, overrunning the work of scribes and eroding the sanctity of scholarship. This moment in history feels eerily familiar today.

If you substitute press with algorithms, scribe with human author, and printer with Artificial Intelligence (‘AI’) developer, the anxiety feels the same. The emergence of artificial intelligence in content creation, law, literature, and the arts isn’t unlike a major tidal shift like what Gutenberg’s invention generated. Where once a manuscript had authority, AI now produces manuscripts with just a trigger word, and in a way is as seamless and voluminous, and disturbingly expeditious. If present-day critics echo Filippo’s despair with updated language and definitions, they do so with good reason. They point out that generative AI, although amplifying, is indiscriminate content. It can generate style without substance, language without wisdom. Just as printers sacrificed humanist care for commercial expediency, AI tools have introduced commentary into the digital commons that can resemble coherence but lacks the thinking of a human. Just as novices were producing degraded Latin, today’s legal scholars quake at the thought of jurisprudence now possibly populated with hallucination and context-less citations.

Generative AI and Fair Use

It is imperative to understand ‘generative’ as a term when it precedes ‘AI’. Generative AI does not simply regurgitate facts but works collaboratively with humans and pushes so further beyond the realm of the conceivable. It can write code that gives solutions to difficult problems, design futuristic buildings, and compose music like Mozart. It helps human creativity, rather than taking away from it, and in doing so, opens new opportunities for investigation and invention.

This rapid-fire AI development has become quite a conundrum in copyright and intellectual property. AI models, of late, can create text, images, and even computer code to rival human works and sometimes best them. This fact has triggered a debate about ownership, fair use, and the very definition of authorship as far as machines are concerned. So, as these models become more sophisticated and their outputs more precise, the once clear demarcation between legitimate inspiration and outright imitation grows into an area of murky water where absolute distinctions are missed. This ambiguity begs the really critical questions about how to protect intellectual property in this fledgling technological terrain. How do we demarcate the AI that learns from existing works and then just generates it with no touch of its own judgement, from one that does nothing but copy from the tools of literature and academics? The legal and moral dealings extend in legions.

The landmark case of Sony Corporation of America v. Universal City Studios in 1984 is quintessential in this regard. The case concerned Universal Studios, which sued Sony over copyright infringement because the Betamax video recorder allowed individuals to record other television shows and was therefore infringing on their copyright. Ultimately, the Supreme Court came down in favour of Sony, holding that the Betamax had ‘substantial non-infringing commercial uses,’ including time-shifting, and the technology itself was not automatically infringing. While Betamax established the concept of ‘substantial non-infringing uses,’ it doesn't provide any sort of blanket excuse from copyright for the use of artificial intelligence. Whether a particular use of AI is within fair use or constitutes infringement will demand case-by-case evaluation against various factors, such as the nature of the use, the type of copyrighted content, the portion used, and its impact on the original work’s market value.

Model Distillation and Case Studies

A huge amount of training data for the models of AI, even up to billions of separate pieces of data, has to be provided during AI model training. Training AI models in academia usually involves massive ingestion of text, including copyrighted books, articles, scholarly papers, and other such research materials. Given the sheer scale of the data, all pieces cannot be manually verified for compliance with copyright laws, and this raises an interesting question about how to proceed on floodgates for unintentional copyright infringement in AI training.

Training AI models with copyrighted material appears to have legal support as ‘transformative use.’ However, it must be ensured that the data is obtained legally. The era of "download first, ask questions later" is seemingly coming to an end. Meta's recent win in a federal copyright case where a judge granted summary judgment in their favour, citing fair use, is making rounds and while prominent authors, including Sarah Silverman, had accused Meta's Llama AI of being trained on their copyrighted works without authorisation, Meta's victory largely hinged on the authors' inability to prove market harm, a key factor in fair-use analysis. Meta successfully argued that its AI, Llama, is a ‘quintessentially transformative’ product, designed to learn from original works for new purposes, rather than to substitute them.

Essentially on one hand, AI companies argue that their systems make fair use of copyrighted material by studying it to learn to create new, transformative content, and that being forced to pay copyright holders for their work could hamstring the growing AI industry but on the other hand, copyright owners say AI companies unlawfully copy their work to generate competing content that threatens their livelihoods. This leads to a legal tussle difficult to navigate.

At the same time, AI developers themselves are not immune to accusations. OpenAI's recent public pronouncement accusing DeepSeek of "inappropriately distilling its models" is a testimony to the prickly undertones in the space. Ironically, while DeepSeek gets publicly hammered, OpenAI itself was in the midst of a messy situation of its own, being accused of copyright infringement by ANI in India.

A question central to this dispute is whether there is any merit to OpenAI's accusations against DeepSeek. To determine this, we must understand what is meant by ‘model distillation’, whereby a much smaller, much more efficient "student" model is trained on the output of a much larger, much more complex "teacher" model. Observably, the student model achieves comparable performance and expectedness to that of the teacher model while using fewer computational resources and storage through training using only the teacher's labels. OpenAI asserts that DeepSeek has obtained this for its own smaller and more efficient models by querying OpenAI's large-scale model repeatedly and then training on its outputs.

AI-IPR Disputes in Asia

The notions of fair use and transformative use in all these cases stand questioned. This act of copying, even if technically different from traditional copyright infringement, mirrors the broader threat AI poses in the long run. The contrast is nowhere starker than in the case of Asia, a region that is actively encouraging technology advancement but also grappling with the legal, ethical, and economic implications of generative AI.

Singapore and Japan have been described as emerging global hubs for AI training, and Singapore particularly came into the limelight because of its amended Copyright Act, which expressly permits computational data analysis (‘CDA’) that allows access to copyrighted works without the need for prior authorisation for legal use in the context of computational analysis. This has created a more favourable jurisdiction for AI developers and researchers creating models based on huge corpora of data (in particular, large language models).

Japan has a similar story. In the past few years, the Japanese government has published two white papers on AI policy, both of which demonstrate an aim to achieve a definitive policy path towards fostering AI innovation and solidifying the ownership of the innovation. However, Japanese jurisdiction is still taking a stance against AI ownership. On 30 January 2025, the Japanese Intellectual Property High Court ruled that, under the current Japanese patent law, AI-generated inventions could not be afforded patent protection. The Court held that the Patent Act allows patents for inventions made by "natural persons," and 'natural persons' both in terms of rights and procedures, thereby highlighting the DABUS case shockwaves.

India, too, is witnessing a growing wave of anxiety on the part of creators. Various publishers in India have argued for royalty schemes when their works are used as training data. This demand alone demonstrates a common desire to push for legislative clarity. The Indian copyright regime does not yet include express protections for AI-generated content or training datasets, and a demand for royalty payment is a clear sign of a growing consensus around economic fairness.

The Way Forward

Given the scale and complexity of these issues, a well-developed multi-layered response to AI-IPR conflict in Asia must be initiated in terms of legal accuracy, technological awareness, and regional appropriateness.

Firstly, a pan-Asian AI-IPR treaty that is similar to the GDPR or Berne Convention must be strongly considered. Such a treaty could introduce baseline norms around fair-use, user consent, and data access, attribution to content creators, and usage of dataset contents. If there were a collaborative framework in place, there would be a greater reduction or prevention of regulatory arbitrage, and it would provide a tool set to bridge creator rights and protections despite border differences.

Secondly, Asian governments could seek to develop automated licensing equivalent to the performance rights organisations in the music space, where creators have the option to opt, and license their works for training purposes. These platforms could enable micro-royalty payments and data use tracking through technologies such as blockchain or digital watermarking. This will create a fair and scalable revenue model for creators.

Thirdly, sharing legal frameworks is key to interoperability. Singapore's CDA exception, Japan's policy inclinations, China's Article 7.2, and India's royalty discussions must be part of the regulatory conversation and dialogue. Varying standards of AI usage in these countries lead to developers ending up in a legal maze wherein they try to strike a balance between strict and lenient regimes. Aligning AI regulations, even to an extent, can lead to better AI operations on the continent. The closer the legal frameworks align, the less uncertainty is created for legal queries, and consequently, cross-border AI development would be simplified.

Finally, the creation of AI-IPR mediation panels could offer domain-specific venues for resolving disputes. These panels, made up of legal scholars, technologists, and industry practitioners, could address the more nuanced and complex layers of claims in a relatively quicker timeframe than traditional courts could. A regional dispute resolution body under organisations like the Association of Southeast Asian Nations (ASEAN) or the World Intellectual Property Organisation (WIPO) could operate in this capacity.

Asia is in a unique position to contour the global norms for AI and IPR. However, the continent will need to engage in inclusive, conscious policy processes that recognise creator rights as foundational, not expendable.

---

This article has been authored by Khushi Mishra, a third-year student at University School of Law and Legal Studies, GGSIPU. It is a part of the RSRR's Blog Series on 'Ideas in Motion: Contemporary Frontiers in Intellectual Property Law’, in collaboration with Ahlawat & Associates.