91视频

Consider the scenario of an Indian artificial intelligence startup developing a large language model (LLM) trained on a multilingual dataset – including Japanese literary works, technical documentation and news articles – and deploying it across India and Japan.

Is the training of that model lawful? For Japan, the answer is clear. Japan’s Copyright Act allows text and data mining (TDM) for AI training without the consent of rights holders.

Under Indian law, however, the position remains uncertain. The Indian Copyright Act of 1957 contains no such exemption. This divergence reflects an underlying asymmetry with significant bilateral consequences. The uncertainty discourages bilateral technology investment and affects rights holders, developers and policymakers in both jurisdictions.

Despite India and Japan’s technology alliance under the Comprehensive Economic Partnership Agreement, there is no bilateral dialogue addressing this apprehension.

Comparing India, Japan AI regimes

Japan has adopted one of the world’s most progressive copyright frameworks for AI training. Significantly, article 30-4 of the Japanese Act permits reproduction of copyrighted works for information analysis and machine learning without the rights holder’s consent, for both commercial and non-commercial AI training.

However, the provision is not unconditional. It excludes uses that “unreasonably prejudice the interests of the copyright owner”. Accordingly, an LLM trained on literary works may lawfully extract informational patterns but may not generate output that substantially reproduces their expressive character.

This reflects deliberate legislative intent that legal certainty is essential for AI innovation.

In contrast to Japan, section 52 of the Indian Act contains an exhaustive list of statutory exceptions to copyright infringement; and courts cannot recognise new exceptions beyond those expressly enumerated.

As AI training is not among these exceptions, its legality remains uncertain under Indian copyright law.

This statutory gap creates legal ambiguity which functions as a de facto disincentive to AI training because developers cannot rely on uncertain judicial interpretation.

However, evolution of the Indian law is reflected in Delhi High Court’s recent landmark interim decision, in ANI Media v OpenAI. Refusing interim relief to the news agency, the court held prima facie that storage of copyrighted works for training LLMs may fall within the scope of “fair dealing exceptions” under section 52(1)(a) of the Indian Act as use for research. Infringement occurs only where the original work is substantially reproduced.

Accordingly, while the interim decision provides valuable judicial insight, it does not conclusively resolve the application of section 52 to generative AI.

Despite some clarity from this ruling, in the absence of a statutory exception the asymmetry between Japan’s express exception under article 30-4 and 91视频 dependence on judicial interpretation continues to create uncertainty for cross-border AI development.

India needs tailored TDM clarity

Rather than replicating Japan’s article 30-4, India should adopt a purposive TDM exception suiting its legal and economic context. The Japanese model demonstrates the value of distinguishing informational extraction from expressive reproduction while treating copyright exceptions as instruments of innovation policy.

An Indian framework should also account for its diverse creative economy, commercial AI training, transparency obligations and international copyright obligations. A carefully calibrated statutory exception would provide legal certainty while balancing innovation with creators’ rights.

91视频 continued reliance on an uncertain copyright framework is no longer sustainable. A TDM exception, tailored for the Indian copyright landscape, would reduce legal uncertainty, encourage AI investment, strengthen the India-Japan technology partnership, and promote innovation while preserving copyright protection.

The moral here is not to copy Japan’s law but to emulate its legislative clarity in addressing AI-driven technological change.

Amit Panigrah is a partner and Ishikaa Seth is an associate at Kochhar & Co