India’s draft AI copyright model may treat everything online as training data, unless one locks it away

This means 'platforms that do not wish their material to be used for model training must introduce robust access controls, such as paywalls, rate limiting, anti-scraping technologies.'

A Department for Promotion of Industry and Internal Trade (DPIIT) working paper on generative AI has suggested that any content that is “lawfully accessed” online could be used to train AI models.

Experts Moneyontrol spoke to say this changes how copyright usually operates on the internet: unless a platform introduces paywalls or other protections, its content effectively becomes available for model training.

“Copyright law does not distinguish between publicly accessible and paywalled content—even material that is freely viewable is subject to protection,” says Pallavi Sondhi, senior associate at Ikigai Law.

The proposed hybrid model, published by DPIIT in its document titled “One Nation One Licence One Payment: Balancing AI Innovation and Copyright' would allow AI developers to rely on “lawfully accessed content”.

This means “platforms that do not wish their material to be used for model training must introduce robust access controls, such as paywalls, rate limiting, anti-scraping technologies.”

The proposal introduces a compulsory licensing mechanism—administered through a proposed Copyright Royalty Collection and Allocation Tribunal (CRCAT)—that would treat accessibility as the key threshold. Under this model, access to content for AI models would hinge on open visibility rather than explicit permission, the paper says.

Jameela Sahiba, Associate Director of The Dialogue, argues that while the model’s intentions are constructive, “its mandatory blanket license creates a uniquely burdensome regime for AI developers operating in India.”

She points out that the model’s compulsory licence would involve reporting “global revenue percentages” annually, government-fixed royalty rates “subject to judicial review,” and a CRCAT structure that still leaves “fragmented representation” for unorganised creators.

According to her, this could force developers into “protracted audits, proportional apportionment disputes, and board governance tangles”.

Sahiba also argues there are “no global precedents where compulsory training licenses for AI have demonstrably translated into meaningful payouts for individual creators,” noting that traditional statutory licence pools have historically yielded “uneven or negligible remuneration” for smaller creators”.

Sondhi, meanwhile, stressed that even for platforms that want to restrict scraping, enforcement will rest on detecting and monitoring violations.

“Ultimately the effectiveness of the platform will depend on the ability of platforms to detect and monitor violations,” Sondhi said.