New Delhi : Esya Centre has released a landmark new report titled “An In-Depth Assessment of the Intersection of AI and Copyright in India,” which finds that India’s fast-growing AI ecosystem is constrained by significant legal ambiguities stemming from an outdated copyright framework.
The report underscores that despite India’s position as home to one of the world’s largest developer communities, rising research output, and a thriving AI startup environment, the absence of clear legal provisions for AI training, text-and-data mining (TDM), and generative AI outputs creates barriers for researchers and enterprises alike.
India’s generative AI ecosystem spans finance, healthcare, education, and governance, yet persistent legal uncertainty over whether training data and AI outputs infringe copyright creates significant operational and investment risk. These ambiguities limit access to high-quality datasets and disproportionately burden startups that lack the legal and financial resources to navigate complex compliance requirements.
Among the first detailed analyses in the Indian context, the report assesses how India’s copyright exceptions under Section 52 fall short of supporting modern AI workflows such as large-scale data ingestion, processing and model training. It also notes that countries like Japan and Singapore have introduced TDM exceptions to enable AI model training, ensuring legal certainty on this front. Despite India’s rapid AI adoption across sectors, progress is constrained by a framework designed for a pre-AI era. The report warns that without timely reforms, India risks losing competitive ground to nations that have modernised their copyright regimes.
Shweta Venkatesan, Fellow, Esya Centre, author of the report, said, “AI is redefining how knowledge, creativity, and innovation come together. India must now ensure that its copyright laws reflect these technological shifts. Our analysis shows that we can protect creators’ interests while enabling the computational uses that are foundational to AI systems. Establishing legal clarity through a wide TDM exception will be central to unlocking India’s full AI potential and ensuring that innovation continues to flourish.”
Key Findings
• The risk of copyright lawsuits related to the AI training process could deter generative AI start-ups from developing innovative products and services. Unlike larger businesses, start-ups may not find it economically feasible to have dedicated legal teams that help them navigate ambiguous and complex regulatory environments.
• Training datasets do not contain copies of works. Rather, they often contain information synthesized from metadata – which is data about data. Metadata is not copyrightable because copyright law protects creative expressions rather than ideas or facts. This principle, termed the “idea-expression dichotomy” has also been reiterated by India’s Supreme Court in RG Anand v Deluxe Films.
• Training datasets may themselves be protected under Indian copyright law. For one, the Copyright Act protects compilations (including computer databases) as literary works, bringing training datasets within its ambit. Additionally, in Eastern Book Company v DB Modak, India’s Supreme Court held that a compilation is protected under copyright if it displays a minimum amount of creativity in its selection and arrangement, along with the application of labour. Thus, training datasets compiled and arranged with some creativity may qualify for protection under the Copyright Act.
• The use of copyrighted works to train AI models is “transformative”, for two reasons. First, AI model training constitutes a non-expressive use of copyrighted work, because it does not convey any protected expression within the work to a third person. Second, some stakeholders noted that generative AI systems are not designed to duplicate any portion of their training data — instead, they merely seek to produce new outputs based on certain salient features or characteristics embodied within it. One can draw an analogy between generative AI training and human learning, noting that both processes involve using pre-existing works to create new and distinct outputs.
• Uncertainty About Application of Fair Dealing to AI Model Training: Generative AI start-ups face uncertainty about the application of the fair dealing exception to AI model training. This is because findings of fair dealing depend on the facts of each individual case, and judicial decisions on this issue exhibit considerable inconsistencies.
• Global Best Practices Provide Clear, Actionable Models: Countries like Japan have enacted broad TDM exceptions to enable AI model training. Introducing a similar exception under Indian copyright law will reduce compliance burdens, attract investment, and enable startups to build globally competitive models.
• Uncertainty Regarding Copyrightability of AI Outputs: There is uncertainty about the copyrightability of AI-generated outputs, as well as the term of protection for such works under Indian copyright law.
This report draws on insights from two expert roundtables. The first, co-hosted by the Artificial Intelligence Knowledge Consortium (AIKC) and the Esya Centre explored “Ingest, Generate, Compete: India’s Roadmap for AI, IP and Competition,” and the second roundtable, convened by the Esya Centre, examined “AI and Copyright: Building Legal Clarity for Start-Ups in India”.
