India does not have an AI governance gap. It has an AI governance fragmentation problem and the two are not the same thing.
Union Minister Ashwini Vaishnaw's recent suggestion that India may need a dedicated AI law has framed the question as a binary: statute or administrative adaptation, EU-style legislation or sectoral flexibility. Both sides are answering the wrong question. The more consequential question is whether the three distinct AI governance regimes India has already built simultaneously, without coordination are coherent with one another. They are not. And neither a well-designed statute nor further institutional adaptation will solve a problem that is architectural in nature.
India currently operates three AI governance frameworks in parallel. The executive framework: the February 2026 amendments to the IT (Intermediary Guidelines and Digital Media Ethics Code) Rules, imposing mandatory obligations on intermediaries dealing with AI-generated content. The judicial framework: the Supreme Court's draft Regulations for Use of Artificial Intelligence in Courts, 2026, a 35-regulation document that amounts to a self-contained AI governance constitution for the judiciary. The advisory framework: the India AI Governance Guidelines released by MeitY in November 2025, a non-binding, principle-based document that explicitly concludes a standalone AI law is not presently necessary. None of these frameworks was designed to interact with the others. None contains a mechanism to resolve conflicts. And the conflicts are already written into the texts.
The Guidelines themselves acknowledge the problem in terms that should end any debate about whether a crisis exists. They state that responsibilities "are distributed across multiple agencies, creating opportunities to enhance cross-sectoral coordination and strategic alignment." The proposed solution an AI Governance Group, a Technology and Policy Expert Committee, an AI Safety Institute is symptomatic of the disorder it seeks to treat. The document proposes solving fragmentation by adding more institutions to an already fragmented landscape, without specifying what authority any of them would have over the two frameworks already in force.
Why Importing the EU AI Act Would Make Things Worse
The EU AI Act has become the reference point around which the Indian debate is organised. Its invocation, on both sides, is analytically misleading.
The EU AI Act is a horizontal risk-classification regime built for a specific context: a supranational bloc with a dedicated cross-border enforcement body, a fundamental rights framework with horizontal effect, and an AI ecosystem dominated by enterprises that are simultaneously developers and deployers. None of these conditions holds in India.
India's AI ecosystem is dominated by deployers, not developers. Most high-risk AI systems operating in India are designed by companies headquartered in the United States, China, or Europe. A risk-classification regime imposed at the deployment stage without reaching developer liability would penalize Indian enterprises for design decisions made abroad. The EU framework addresses this through extraterritorial provisions backed by enormous market leverage. India does not have equivalent leverage over the foundational model developers whose systems its deployers use.
The EU AI Act was finalized in 2024 after a four-year legislative process and is still being operationalized. AI governance literature is littered with the wreckage of frameworks that governed last decade's technology. India would be copying a model that may already be misaligned with AI systems by the time any comparable statute is enacted and operationalized here. India's governance problem is not that it lacks Europe's answer. It is that it lacks a coherent version of its own question.
The Guidelines are correct that India should not enact an AI Act modeled on the EU framework. They are not correct that the absence of such a statute means India's governance architecture is adequate.
The Definitional Fault Line Already Written Into Law
The most concrete evidence of governance fragmentation is textual. A comparison no commentary has yet performed reveals a definitional incoherence at the foundation of the two enforceable frameworks.
The February 2026 amendments define "synthetically generated information" as content algorithmically created in a manner that appears real and depicts or portrays an individual or event in a way indistinguishable from a natural person or real-world occurrence. The definition is anchored in deception: what matters is whether the content misleads a viewer about its origin.
The draft Regulations issued by the Supreme Court have a different definition of "synthetic data or synthetic information". The content of Regulation 3(1)(zi) refers to content that "mimics statistical properties of real data without being directly collected from real-world events or persons. It doesn't have to look like genuine content. No requirement that it be of a real person. The definition is based on the epistemic source of the data, that is, whether it was produced by a model or by observation in the real world.
The same content, created by the same technology, on the same day, but different legal outcomes from these definitions. By design, a large language model-generated legal brief shares the statistical characteristics of legal argumentation. It does not portray a true person in a way that is indistinguishable from reality. Under the IT Rules, it is not even included in the definition. It is synthetic information that is required to be disclosed to the court under Regulation 43(3) in accordance with the Supreme Court's rules. An AI-assisted, advocate-drafted document is subject to an obligation that is not subject to in the other — simultaneously, under both.
This is not a drafting error that is to be corrected in revision. It captures a true sense of divergent views on the legal value of AI-generated content from two institutions that were not in communication.
The Guidelines' Self-Contradiction
The Guidelines' conclusion that a separate AI law is not presently necessary deserves scrutiny rather than acceptance. The document states that "many of the risks emerging from AI can be addressed through existing laws." Two paragraphs later, it states there is "an urgent need to conduct a comprehensive review of relevant laws to identify regulatory gaps." These positions are in direct tension. If existing laws are adequate, what do the gaps require? If the gaps are real, what fills them absent new legislation? The document does not resolve this because doing so would force a choice between two positions it is unwilling to abandon simultaneously.
There is a further problem. The Guidelines' seven sutras include "Innovation over Restraint" the principle that "all other things being equal, responsible innovation should be prioritised over cautionary restraint." This principle has migrated, verbatim, into the Supreme Court's draft Regulations. Regulation 17, titled Innovation over Restraint, creates a presumption in favour of AI adoption in courts. Courts have never before operated under a pro-innovation presumption. The Bangalore Principles of Judicial Conduct which the draft Regulations themselves affirm continue to govern contain no such principle. The importation of a MeitY policy sutra into a judicial governance framework, without any analysis of its constitutional consistency with the adjudicative role of courts, is precisely what uncoordinated governance produces.
The Constitutional Incoherence
The governance fragmentation has constitutional dimensions neither framework addresses.
The Supreme Court's draft Regulations establish an Apex Body under Regulation 22 to govern AI deployment across India's entire judicial system. Among its permanent ex officio members is "an officer not below the rank of Joint Secretary to the Government of India in the Ministry of Electronics and Information Technology." An executive official, appointed by and accountable to the Central Government, sits permanently on the highest policymaking body for AI in the judiciary. The Supreme Court has treated judicial independence from executive interference as part of the basic structure of the Constitution. The draft Regulations create a structural mechanism through which the executive participates in approving AI systems for judicial deployment and setting mandatory standards for AI governance across all courts. Whether this arrangement is constitutionally permissible is a serious question. The regulations do not ask it.
The Guidelines' proposed AI Governance Group has no mechanism to issue directions to this Apex Body. The "whole-of-government" coordination architecture proposed by the executive explicitly excludes the judiciary's own governance framework — India's most institutionally sophisticated AI governance initiative to date. The executive and the judiciary are building parallel governance universes with no interface between them.
The Rights Architecture That Does Not Exist
Both enforceable frameworks impose obligations on institutions without creating corresponding rights for the individuals those institutions affect.
The IT Rules 2026 introduce due diligence requirements for intermediaries: labelling, timelines for takedowns of 3 hours, and preservation of evidence. They do not establish a legal obligation on AI to prevent harm to individuals. A grievance officer is available for a person who is misrepresented in a deepfake. In the event that the grievance mechanism is not effective, they do not have a cause of action against the person who directed the AI, the developer of the AI model used, or the platform. Enforcement is reliant on the compliance of the intermediaries, which was tightened up in the 2026 amendments, but structurally remained the same.
The Supreme Court draft regulations create the audit registers of institutions, databases of incidents, transparency reports, but the accountability does not go from the litigant to the Apex Body, it does not go downwards, it goes upwards. Under Regulation 8, officers are not allowed to use "the occurrence of hallucination" as an excuse to evade responsibility for a bad decision. It's a good rule of officer responsibility. It would be of no use to a litigant who wishes to attack an adverse decision on the basis that the adverse decision was caused by hallucination, but who cannot get any remedy unless the usage was "prohibited" under the strict conditions of Regulation 20.
The Guidelines suggest a 'graded liability system'. This is the correct structure. It is mentioned in a non-binding document. The enforceable frameworks do not include individual remedies, but rather institutional obligations. The non-binding document does not include the rights architecture with enforcement. The inversion is exact.
What Meta-Governance Actually Requires
India does not need an EU-model AI Act. But it needs something none of the existing instruments provide: a meta-governance framework that resolves foundational conflicts rather than adding institutions to an uncoordinated landscape.
This requires four things. Common definitional foundations for AI-generated content across regulatory domains the IT Rules and the Supreme Court Regulations proceed on incompatible theories, and only an authoritative instrument can choose between them. A constitutional allocation of AI governance authority between the branches the Apex Body's composition question will not resolve itself, and leaving it for litigation is a choice, not an omission. An affirmative rights framework for AI-affected persons when India's frameworks are tested in constitutional litigation, and the Regulation 8 paradox is a ready vehicle, the absence of individual remedies will be the first problem courts face. And a coordination mechanism with genuine authority the Guidelines' proposed Governance Group has no jurisdiction over the Supreme Court's Apex Body, no power over the RBI's separate FREE-AI framework for the financial sector, and no mechanism to resolve the conflicts already in force. Adding a fourth institution to three uncoordinated ones is not coordination.
None of these objectives requires comprehensive horizontal legislation. None can be achieved without deliberate architectural choices that India has so far declined to make.
No statute solves an architectural problem. But the claim that distributed institutional adaptation, proceeding on incompatible foundations, will eventually converge into coherent AI governance is not a policy position. It is optimism without a mechanism.
India's AI governance crisis is not that it lacks a law. It is that it lacks the architectural honesty to acknowledge what three simultaneous, uncoordinated, and contradictory regulatory regimes actually are and to make the foundational choices required to build something coherent from them before those contradictions become the subject of litigation that forces the resolution that policymaking chose to defer.
Author is a fifth-year Law student at National Law University Odisha. Views are personal.