MeitY AI Regulatory Advisories and Synthetic Media Watermarking: Intermediary Due Diligence under IT Rules 2021

I. I. The Regulatory Shift from Laissez-Faire to Executive Intervention

India's approach to artificial intelligence regulation experienced a dramatic paradigm shift between 2023 and 2024. For years, the Union Government adopted an explicitly market-friendly, innovation-centric posture, stating that India would not prematurely legislate AI to avoid throttling startup dynamism. However, the sudden explosion of viral deepfakes targeting high-profile political figures during national elections, combined with pervasive algorithmic bias and election interference concerns, prompted immediate executive intervention.

Acting through the Ministry of Electronics and Information Technology (MeitY), the Union Government utilized its executive powers under the Information Technology Act, 2000, and the Information Technology (Intermediary Guidelines and Digital Media Ethics Code) Rules, 2021 (IT Rules 2021). Rather than waiting for a comprehensive Digital India Act, MeitY issued a series of binding regulatory advisories commanding internet intermediaries, social media platforms, and AI developers to implement immediate compliance safeguards.

This regulatory evolution reached its zenith with the landmark MeitY Advisories of November 2023 and March 2024. These regulatory pronouncements imposed unprecedented obligations upon technology platforms, establishing that the deployment of under-tested or biased models, and the failure to cryptographically watermark synthetic media, directly jeopardizes an intermediary's statutory safe harbor under Section 79 of the IT Act.

II. II. Deconstructing the March 2024 MeitY AI Advisory

On 1 March 2024, MeitY issued a sweeping advisory to all intermediaries and generative AI platforms operating within Indian digital borders. While an initial draft suggested that platforms must obtain explicit government permission before deploying 'untested' or 'unreliable' AI models to Indian users, an amended clarification issued on 15 March 2024 structured the mandate around three core compliance pillars.

First, intermediaries must ensure that the use of artificial intelligence models, Large Language Models, generative AI, or software algorithms does not permit any bias, discrimination, or threat to the integrity of the electoral process. Platforms are commanded to rigorously test their models prior to public release to eliminate discriminatory outputs across religious, caste, linguistic, and gender lines.

Second, platforms hosting or facilitating generative AI must clearly inform users through terms of service and user agreements that computer resources must not be used to host, display, upload, modify, or transmit content that is patently false, untrue, or defamatory. Third, and most consequentially, the advisory established mandatory watermarking and labeling protocols for all synthetically generated or modified audio, visual, and multimedia content.

III. III. Mandatory Watermarking, Metadata Labeling, and C2PA Standards

Under the MeitY framework, whenever an artificial intelligence system generates, modifies, or alters an image, video, or audio recording, the deploying intermediary must affix a clear, permanent, and identifiable label or watermark indicating that the media is synthetic. This requirement operates across two technical levels: perceptible visual/auditory watermarks and imperceptible cryptographic metadata.

Perceptible watermarks must be prominently visible or audible to ordinary users, clearly stating 'AI-Generated' or 'Synthetically Modified'. However, because malicious actors can crop or filter visual labels, MeitY mandates the integration of persistent cryptographic metadata. This requires platforms to adopt global open standards such as the Coalition for Content Provenance and Authenticity (C2PA) framework.

Under C2PA standards, digital media files are embedded with cryptographically signed provenance manifests. These manifests record the identity of the generating model, timestamp, hash values, and tool configurations. If an individual attempts to alter the media or strip the metadata, the cryptographic chain of custody is invalidated, providing forensic evidence of tampering under Section 63 of the Bharatiya Sakshya Adhiniyam 2023.

IV. IV. Intermediary Safe Harbor under Section 79 and the Due Diligence Nexus

The legal teeth behind MeitY's advisories lie in Section 79 of the Information Technology Act, 2000. Under Section 79(1), intermediaries are granted immunity from third-party civil and criminal liability for content hosted or transmitted through their platforms. However, Section 79(2)(c) conditions this safe harbor on the intermediary observing strict 'due diligence' prescribed by the Central Government under the IT Rules 2021.

MeitY's regulatory strategy bridges executive advisories and statutory due diligence. Rule 3(1)(b) of the IT Rules 2021 enumerates prohibited categories of content, including misinformation, impersonation, and content threatening public order. MeitY's advisories clarify that deploying generative AI tools that predictably generate prohibited content—without watermarking or filtering guardrails—constitutes a failure to observe statutory due diligence under Rule 3.

Consequently, an intermediary that fails to implement synthetic media labeling faces the immediate revocation of its Section 79 immunity. Once safe harbor is stripped, the platform stands in the shoes of a publisher, facing direct co-conspirator and abetment liability under Section 356 BNS (defamation), Section 336 BNS (forgery), and Section 66D IT Act (cheating by personation).

V. V. Enterprise Compliance Architecture: Audit Workflows and MLOps Guardrails

To maintain full statutory compliance with MeitY mandates, technology enterprises operating in India must institutionalize a multi-layered compliance workflow. Machine learning engineering pipelines must incorporate automated pre-release bias stress-testing, utilizing standardized adversarial datasets to measure output distribution skew before model checkpoints are signed off.

Product teams must integrate automated watermarking into the inference pipeline itself. In image generation architectures (such as Stable Diffusion or Midjourney APIs), imperceptible watermarking algorithms (e.g., SynthID or cryptographic steganography) must be injected into the frequency domain of pixel arrays, ensuring the watermark survives compression, cropping, and screenshotting.

In addition, enterprise terms of service, End-User License Agreements (EULAs), and developer API agreements must be updated to include explicit prohibitions against stripping metadata, utilizing models for electoral disinformation, or deploying unwatermarked outputs in public communications. Platforms must maintain searchable internal registries documenting model evaluation metrics and audit logs for regulatory inspection.

VI. VI. Chambers Representation in Regulatory Inquiries and Intermediary Audits

Advocate Ritik Khurana Chambers advises global social media platforms, domestic generative AI startups, and cloud infrastructure providers on MeitY regulatory compliance, intermediary liability defense, and platform governance. We represent technology companies in formal regulatory proceedings, drafting statutory responses to MeitY inquiry notices and CERT-In compliance directives.

Our team conducts comprehensive Intermediary Due Diligence Audits, verifying that platform terms of service, automated watermarking systems, and grievance redressal mechanisms conform strictly to the IT Rules 2021 and evolving administrative advisories. In high-stakes litigation where platform safe harbor is challenged before the High Court, our chambers delivers formidable trial advocacy to protect corporate immunity and uphold constitutional free expression.

Navigating India's dynamic AI regulatory landscape requires technology leadership to harmonize engineering agility with rigorous, proactive statutory compliance.