CHENNAI: Even as the country accelerates investments in domestic AI capabilities, India's push to build sovereign artificial intelligence systems must confront the complex governance, legal, and accountability challenges, according to a policy white paper released by the Office of the Principal Scientific Adviser to the Government of India.
The document, ‘Advancing Indigenous Foundation Models’, underscores that while India has begun building an ecosystem for home-grown foundation models, the rapid expansion of generative AI has raised difficult questions around regulation, copyright, model accountability, and the governance of synthetic content.
Foundation models, which are large AI systems trained on massive datasets that power applications ranging from language processing to healthcare diagnostics, are increasingly becoming the backbone of digital economies. But the paper warns that decisions taken during model design and training can influence thousands of downstream applications across sectors.
"Choices made at the model's design and training stage can shape performance and risks across many downstream uses," the paper notes, highlighting the cascading impact these systems can have across industries.
A key concern flagged in the report is the lack of clarity around accountability across the AI value chain. In many cases, models are developed by one organisation and deployed by another, making it difficult to determine responsibility when AI systems generate biased, inaccurate, or harmful outcomes.
The report argues that accountability cannot be confined to application developers alone since upstream design decisions by model creators often influence the behaviour of deployed systems.
Experts say this emerging governance gap is one of the most complex policy challenges facing AI regulation globally.
"Foundation models are no longer just research artefacts; they are becoming critical digital infrastructure. That means responsibility cannot sit only with the last developer in the chain. Governance frameworks must recognise the layered nature of AI development," an AI researcher, who has worked on large-scale machine learning systems from the IIT Madras told DT Next.
The white paper also highlights unresolved copyright issues surrounding the data used to train generative AI models. As these systems are trained on vast collections of publicly available material, concerns have grown over whether such practices infringe on intellectual property rights.
To address this, the government is examining a hybrid framework that would allow AI developers to train models on lawfully accessed data while requiring royalty payments once AI tools are commercially deployed.
Another challenge identified is the regulation of synthetically generated content, including AI-generated images, audio, and video. Proposed rules would require such content to carry visible labels and embedded identifiers to prevent misuse and improve transparency on digital platforms.
Industry experts say such measures are necessary but must be implemented carefully to avoid stifling innovation.
"Regulation is inevitable as generative AI scales, but the challenge is to design frameworks that protect users without slowing technological development," said R Vishnu Gopalan, an artificial intelligence expert based in Bengaluru.
The report also calls for India-specific benchmarking standards to evaluate the performance, fairness and reliability of AI systems across the country's languages and social contexts, warning that global benchmarks often fail to capture the realities of India's linguistic diversity.