Naman Rathore is a third-year student at National Law Institute University, Bhopal
Introduction
The draft notification proposing amendments to the Information Technology (Intermediary Guidelines and Digital Media Ethics Code) Rules, 2021 (“Intermediary Rules”) issued by the Ministry of Electronics and Information Technology (“MeitY”) is India’s first definitive move toward regulating generative AI and synthetic media under the framework of the Information Technology Act, 2000 (IT Act). These draft amendments emerge against the backdrop of soaring social outrage and judicial combat with the challenges posed by deepfake and AI-generated content. While the proposed amendment brings out a new definition of ‘Synthetically generated information’(“SGI”), it also expands the due diligence requirements of intermediaries, along with new obligations for significant Social Media Intermediaries. It also comes with protection for the removal of harmful synthetic content, effectively codifying the good-faith takedown principle. While the proposed amendment touches upon several areas posing interpretational difficulty, the present blog shall focus on the critical ambiguities within the new Rule 3(3), which threatens to undermine the regulation’s effectiveness.
The Core Mandate: Expanded Due Diligence and the 10% Rule
The proposed amendment centres around a new rule 3(3), which significantly expands the due diligence requirements for intermediaries. This rule mandates that intermediaries offering computer resources that may enable, permit or facilitate the creation, generation, modification or alteration of information as synthetically generated information to ensure that every such information is prominently labelled or embedded with a permanent unique metadata or identifier. This label, metadata or identifier must be visibly displayed or made audible in a prominent manner on or within that synthetically generated information, covering at least ten percent of the surface area of the visual display or, in the case of audio content, during the initial ten percent of its duration. The intermediaries are not allowed to enable the modification, suppression or removal of such label, permanent unique metadata or identifier.
The ‘Intermediary’ Trap for AI Developers
The proposed mandatory duty under this rule to embed and maintain labels brings out a fundamental question, namely, who the rule is intending to regulate and what organisations are considered “intermediaries” under the statute? Under Section 2(1)(w) of the IT Act, an “intermediary is defined as any person who receives, stores, or transmits electronic records on behalf of another.” This definition seems archaic, as it was written long before the launch of generative AI models.
The question lies in whether generative platforms or AI model developers fit within this traditional definition. These platforms neither host nor transmit third-party content in the traditional sense, but only generate new content. This distinction was highlighted by the Delhi High Court in the case of Google LLC v. DRS Logistics Pvt. Ltd. (2023 SCC Online Del 4809), where the court held that intermediary status must be determined contextually, depending on whether the entity functions merely as a passive conduit or assumes an active role in content curation or modification. Applying this logic, most developers of foundational AI systems are likely to be classified as technology developers or service providers, operating outside the IT Act’s established intermediary framework. The present drafting channels the regulatory pressure primarily on secondary enablers of synthetic media dissemination, such as the social media platforms. It, however, circumvents the foundational model developers who are the primary creators of the underlying generative technology, holding the driving seat over the content’s technical creation. This failure to draw the crucial legal distinction between content hosts and content generators provides a loophole to the very entities responsible for the underlying generative technology to remain unaffected by the new due diligence requirements.
The Ambiguity of ‘Modification’
The inclusion of the term ‘modification’ also introduces a significant point of confusion. The rule suggests that facilitating the “modification or alteration of information as synthetically generated information” triggers the full labelling obligation, thereby raising the question of whether even minor visual changes, such as light retouching, background blur, or noise reduction, now qualify as ‘modification’ to trigger the mandatory 10% label. The present phrasing fails to differentiate between material, deceptive synthetic changes like swapping a person’s face and other harmless corrections that are often performed by non-generative tools. This lack of clarity is particularly problematic in the present scenario, as many generative platforms have incorporated basic AI-powered editing tools.
Serious issues are posed due to this ambiguity as it could lead to needless compliance burdens, inconsistent enforcement and compelling organisations to mark normal, non-deceptive modifications. Content creators who use minor AI-assisted editing for legal commercial purposes could face significant obstacles. The focus of the proposed law, which should have been on deceptive, high-impact synthetic media, may unintentionally affect legitimate content production due to this ambiguous term.
Technical Impossibility of the 10% Rule
The requirement to display a permanent label covering at least 10% of the visual area or the initial 10% of the audio duration presents significant technical and user experience challenges. For the visual content, many common content formats such as memes and dynamic graphics are inherently ill-suited to such large, rigid overlays as it would destroy the context or readability of the content itself. Similarly, for Audio content, a continuous audible disclosure would quickly compromise user experience and accessibility, potentially driving users away from platforms that implement it. Even the platforms that integrate AI models across multiple services would need to tag every single generated output, which demands substantial technical, user experience, and compliance resources to manage high-volume outputs across diverse content formats. Furthermore, the mandate for visible labelling in every case overlooks user-friendly technical solutions.
Considering the existence of many content types and their rapid evolution, even though the proposed amendment was sought to bring greater transparency but through such rigid quantitative labelling thresholds it could create overwhelming compliance impracticalities and inconsistent enforcement.
Conclusion and Suggestions
The proposed amendments by MeitY mark a significant step toward proactive regulation of Synthetic and AI-generated content. However, the draft for Rule 3(3) risks becoming ineffective because of legal and technical overreach. The law’s reliance on the archaic definition of an ‘Intermediary’ provides a loophole for core AI developers, while the vague scope of ‘modification’ burdens platforms with tagging minor, non-deceptive edits. Therefore, the rigid 10% rule and the operational challenges it presents for diverse content, especially text and dynamic displays, threaten to make compliance impractical.
To ensure effective regulation, MeitY needs to redefine accountability. It needs to clarify the ‘Intermediary’ definition in a way that it explicitly captures foundational AI model providers at the source of creation. It also needs to remove the ambiguity surrounding modification by distinguishing between material, deceptive synthetic changes and innocuous content correction. There is also a requirement to replace the rigid 10% metric with a technology-neutral requirement centred on non-removable metadata and industry-aligned provenance standards, reserving visible labels only for high-risk, deceptive content. This balanced approach will safeguard users without stifling the essential innovation of India’s AI ecosystem.
From the draft rules, it is clear that the majority of the focus has been on regulating the Synthetically generated information through intermediaries. Instead of it, the focus must be placed on actual generators, publishers, or bad actors responsible for such content. Since the proposed labelling and identifier may hamper the user experience, adopting a risk-based app approach, based on factors such as the nature, duration, and type of SGI, for labelling and embedding identifiers in SGI, could be a better option.