Sovereign AI vs. global giants: India’s struggle to tax the algorithms of OpenAI and Google

As of late 2025, India is attempting its most ambitious overhaul of copyright law in decades, driven by the rise of generative AI and mounting disputes between content creators and AI developers.

In December 2025, the Department for Promotion of Industry and Internal Trade (DPIIT) released a landmark working paper proposing a statutory system titled “One Nation, One Licence, One Payment.” The framework seeks to resolve the legal grey zone around AI training and copyright by introducing a mandatory blanket licence for AI developers. Under the proposal, AI systems would be allowed to train on all lawfully available copyrighted works without prior permission, but once the system is commercially deployed, developers would be required to pay a percentage of their revenue into a central royalty pool.

The paper further recommends retroactive royalty payments for AI models already trained and commercially successful. These funds would be collected and distributed by a proposed non-profit umbrella entity called the Copyright Royalties Collective for AI Training (CRCAT).

At the same time, India’s copyright regime remains unsettled on the ownership of AI-generated content. Under the Copyright Act, 1957, authorship is tied to human creativity. Purely AI-generated works without human intervention remain generally uncopyrightable, while works produced using AI tools may qualify for protection if a human provides sufficient “skill and judgment.” This uncertainty was highlighted by the RAGHAV AI case, where the Copyright Office initially granted co-authorship to an AI tool before withdrawing it, reaffirming that an “author” must be a natural person under Section 2(d).

The use of copyrighted content for AI training is also under active judicial scrutiny. In the ongoing ANI v. OpenAI litigation before the Delhi High Court, the news agency has alleged unauthorised use of its content for model training. The government’s 2025 position is that commercial-scale AI training does not fall under India’s limited “fair dealing” exceptions in Section 52, making licensing or statutory authorisation necessary.

Alongside DPIIT’s initiative, the Ministry of Electronics and IT (MeitY) issued the India AI Governance Guidelines in November 2025, and the IndiaAI Mission, backed by a ₹10,000-crore budget, aims to promote a safe and trusted AI ecosystem.

Yet legal experts warn that the copyright proposal, while ambitious, may face serious constitutional, practical, and economic obstacles.

Design risks and administrative friction

Bharadwaj Jaishankar, Partner at CMS INDUSLAW, described the proposal as an important attempt to modernise copyright law but warned that its structure may inadvertently create bureaucratic obstacles.

“The framework proposed by the DPIIT represents an important attempt to address emerging copyright issues arising from AI training in India. While its intent to safeguard creators and enable the growth of AI is commendable, the approach may pose risks. The proposed model envisages revenue sharing, the creation of statutory licensing bodies, and state-determined royalty structures. Although these mechanisms are not entirely novel, their application in the context of AI and copyright remains underexplored. This could inadvertently create bureaucratic hurdles rather than creating a level playing field,” Jaishankar said.

He flagged the lack of clarity around how copyrighted works would be categorised and valued.

“One concern is the lack of clarity around procedures for categorizing and valuing copyrighted works. The inability of creators to negotiate license terms or influence how, by whom, and to what extent their works are used only exacerbates the problem. Greater transparency with respect to compensation mechanisms and valuation methodologies could help address these issues,” he added.

Constitutional fault lines: royalty setting and separation of powers

Sonam Chandwani at KS Legal describes the proposed royalty structure as the framework’s most constitutionally fragile component.

“India’s proposed AI copyright framework, particularly the attempt to fix or influence royalty rates through a government-linked collective, is the most constitutionally vulnerable limb of the proposal,” Chandwani said.

Indian copyright jurisprudence has consistently held that price or royalty determination without clear legislative standards risks excessive delegation and violates separation of powers principles, especially when executive authorities displace market or adjudicatory functions.

While the proposal’s requirement of “lawful access” for AI training may survive judicial scrutiny as a neutral safeguard, Sonam cautioned that mandatory revenue sharing raises serious constitutional and enforceability concerns.

“Courts are unlikely to impose ex-post revenue obligations on foreign trained models merely because outputs are monetised in India, absent express extraterritorial intent and reciprocal enforcement frameworks,” Chandwani said.

According to her, the framework risks burdening domestic AI developers with compliance and uncertainty, while global AI firms remain structurally insulated, ultimately chilling Indian innovation without delivering meaningful compensation to creators.

Enforcement dilemma: global training, local law

Advocate Prashant Mali echoed the enforcement challenge. “Deeming it as highly unrealistic, he says that Indian courts assert jurisdiction over use and impact in India, but training happens extraterritorially and lacks a clear act of infringement within India.

He added that while India can influence the economics of AI, the limits of jurisdiction remain stark.

“Only Parliament can command it, and only markets, not courts, can police foreign training today,” he said.

This creates a structural asymmetry: foreign AI companies can train models offshore and sell them into India while largely bypassing the proposed compliance regime, whereas Indian developers remain directly exposed to regulatory burdens.

Innovation at risk: over-prescription versus growth

Advocate Siddharth Chandrashekhar characterised the proposal as among the world’s most interventionist AI copyright regimes.

“This prima facie appears to be one of the world’s most interventionist Copyright-AI proposals, pivoting away from text-and-data-mining exceptions and instead imposing a revenue-sharing regime tied to commercial deployment,” he said.

While the framework aims to engineer fairness through statutory collectives, layered royalties, and revenue sharing, Siddharth warned that it risks becoming a new arena of government price control.

“Tomorrow’s AI entrepreneurs may need compliance calendars longer than their model training scripts. For a country that wants to be the world’s AI playground, adding bureaucratic red tape invites constitutional challenges on proportionality, arbitrariness, and excessive delegation,” he said.

Without robust monitoring capacity, the system could become “symbolic theatre,” he added, punishing only those firms operating within India’s reach while global giants sidestep the compliance maze.

Legal vulnerability: statutory limits

Both Chandwani and Chandrashekhar point to a core legal vulnerability: while the Copyright Act allows statutory licensing under Sections 31 and 31D, courts have repeatedly warned against executive price-fixing without explicit parliamentary guidance.

“Policy can guide, Parliament must decide, babus can’t play Copyright Board,” Chandrashekhar remarked.

If AI royalty rates are determined by ministries or regulators without formal legislative amendments, the framework risks being struck down for violating separation of powers.

Potential upsides for creators

Despite the risks, Siddharth acknowledged potential advantages. A blanket licensing system could offer predictable compensation for creators, reduce transaction friction, and streamline compliance through a unified body.

However, the broader concern remains that the architecture, as currently conceived, may reward global incumbents while burdening homegrown AI labs, undermining India’s ambition to become an AI powerhouse.

“A light-touch, exception-based system with judicial calibration would certainly serve India’s innovation arc better,” Siddharth concluded.