India’s startup ecosystem has grown into one of the most dynamic in the world, yet founders and investors continue to grapple with a tax provision that was never designed with them in mind. Section 79 of the Income Tax Act, 1961 [1] is an anti-abuse rule, written to prevent profitable companies from buying loss-making shells purely to harvest accumulated tax losses. In practice, however, it frequently punishes genuine business reorganisations — investor funding rounds, internal group restructurings, and founder share transfers — that have nothing to do with tax avoidance. For closely held startups, which almost by definition undergo regular changes in shareholding as they move through successive rounds of venture capital, Section 79 can silently wipe out years of accumulated losses, destroying the very deferred tax asset that makes a young company attractive to future investors. This article examines what Section 79 provides verbatim, how it has been amended to accommodate startups, what the courts have said about it, and where the provision still falls dangerously short of protecting legitimate commercial restructuring.

What Section 79 of the Income Tax Act, 1961 Actually Says

The operative text of Section 79(1) of the Income Tax Act, 1961 reads as follows:

“Notwithstanding anything contained in this Chapter, where a change in shareholding has taken place during the previous year in the case of a company, not being a company in which the public are substantially interested, no loss incurred in any year prior to the previous year shall be carried forward and set off against the income of the previous year, unless on the last day of the previous year, the shares of the company carrying not less than fifty-one per cent of the voting power were beneficially held by persons who beneficially held shares of the company carrying not less than fifty-one per cent of the voting power on the last day of the year or years in which the loss was incurred.” [1]

The provision applies exclusively to closely held companies — those not substantially owned by the public. Its single operative test is whether persons who beneficially held shares carrying at least 51% of the voting power on the last day of the year in which the loss was incurred continue to hold that same 51% on the last day of the year in which set-off is claimed. If that continuity is broken, the accumulated business losses simply lapse. Critically, the section uses the phrase “beneficially held” and not merely “held,” a distinction that the courts have spent decades unpacking.

The section was introduced pursuant to the recommendations of the Taxation Inquiry Commission — the Mathai Commission — of 1953–54. Its stated purpose was to curb the practice of profitable enterprises acquiring loss-making undertakings solely to use accumulated tax losses to offset their own profits, what is colloquially known as “trafficking in losses.” [2]

The Startup Carve-Out: Finance Act 2017 and the Section 80-IAC Proviso

Recognising that startups raise capital through multiple rounds of dilutive equity funding and that founders frequently exit or reduce their holdings as the company matures, the Finance Act, 2017 inserted a specific proviso to Section 79(1). The proviso provides that even if the 51% continuity condition is not satisfied, an eligible startup as referred to in Section 80-IAC of the Act may still carry forward and set off its losses, provided that all shareholders who held shares carrying voting power on the last day of the year in which the loss was incurred continue to hold those shares on the last day of the year in which set-off is sought. [3] This exemption is available only for losses incurred during the seven years beginning from the year of incorporation.

The Finance (No. 2) Act, 2019, effective from 1 April 2020, substituted the entire Section 79 and introduced additional statutory exceptions — changes in shareholding arising from death of a shareholder, from a gift to a relative, from the amalgamation or demerger of a foreign parent company subject to 51% continuity of that foreign company’s shareholders, from a resolution plan approved under the Insolvency and Bankruptcy Code or under Section 242 of the Companies Act, 2013, and from the strategic disinvestment of a public sector company. [1]

The startup carve-out sounds generous on its face. But the condition that all original shareholders must continue to hold their shares is, in practice, far more restrictive than the general 51% test applicable to other companies. For ordinary closely held companies, new investors may acquire up to 49% of the voting power without triggering Section 79. For eligible startups, a single original shareholder who exits — even a minor angel investor who held 1% — can theoretically put the entire accumulated loss at risk. This asymmetry reflects a legislative choice to ensure that the startup exemption covers only situations where the founding group remains completely intact, but the real-world impact is that it excludes precisely the kind of early investor churn that is normal and healthy in startup financing.

The Concept of Beneficial Ownership: Where Courts Have Disagreed

The single most litigated question under Section 79 is what it means for shares to be “beneficially held.” Two High Courts have reached diametrically opposite conclusions on this question, and the resulting uncertainty continues to affect corporate restructurings across the country.

The Karnataka High Court addressed this in CIT v. AMCO Power Systems Ltd. [4], decided in 2015. In that case, all shares of AMCO Power Systems were originally held by AMCO Batteries Ltd. (“ABL”). ABL transferred a portion of its shares to its wholly owned subsidiary, AMCO Properties and Investments Ltd. (“APIL”), and later transferred 49% of its remaining shares to Tractors and Farm Equipments Limited (“TAFE”), an unrelated party. At that point, ABL directly held only 6%, APIL held 45%, and TAFE held 49%. The Revenue disallowed the carry forward of losses on the ground that ABL no longer held 51% of the voting power directly. The Karnataka High Court disagreed. It held that since ABL was the holding company of APIL and controlled APIL’s Board entirely, ABL effectively exercised voting power over APIL’s 45% stake, and together ABL and APIL controlled 51%, all ultimately under ABL’s direction. The Court ruled that Section 79 “speaks of 51% voting power” and that the purpose of the provision is to prevent losses from being misused by a new owner — a purpose entirely absent where control never left the ABL group. [4]

The Delhi High Court took a completely contrary view in Yum Restaurants (India) Pvt. Ltd. v. ITO [5], decided in January 2016. There, 100% of Yum India’s shares were transferred from Yum Asia Pte. Ltd. to Yum Restaurants International (Singapore) Pte. Ltd., with the ultimate parent throughout being Yum! Brands USA. The taxpayer argued that the ultimate beneficial owner was always Yum USA and that no real change in beneficial ownership had taken place. The Delhi High Court rejected this, holding that Yum Asia and Yum Singapore were distinct legal entities and that there was no agreement or arrangement on record demonstrating that the beneficial owner of the shares was Yum USA rather than the immediate holding entity. The Court found that there was “indeed a change of ownership of 100% shares of Yum India from Yum Asia to Yum Singapore, both of which were distinct entities” and that the “question of piercing the veil at the instance of Yum India does not arise.” [5] Section 79 was held applicable and Yum India was denied set-off of its accumulated losses.

These two cases represent the sharpest fault line in Section 79 jurisprudence. The Karnataka court applied a substance-over-form approach, treating consolidated group control as determinative. The Delhi court demanded explicit contractual or documentary evidence of beneficial ownership in an entity beyond the registered shareholder. Since neither decision has been overturned by the Supreme Court, taxpayers remain subject to different standards depending on the jurisdiction in which they are assessed — a situation that generates unpredictability in precisely the kind of cross-border group reorganisations that Indian startups most commonly undertake.

The Supreme Court’s Foundational Position

The Supreme Court laid down the foundational interpretation of Section 79 in Commissioner of Income Tax, Bombay v. Italindia Cotton Co. (P) Ltd. [6], decided on 5 September 1988. The question was whether the two conditions that save a company from Section 79 — that 51% of voting power continues to be beneficially held by the same persons, or that the change in shareholding was not effected with a view to avoiding tax — operate cumulatively or in the alternative. The Supreme Court held that they operate in the alternative. It stated that the benefit is available “notwithstanding the change in shareholding in the previous year, if shares representing not less than 51% of the voting power remain beneficially held by the same persons on the relevant dates” and equally available if “the change was not effected with a view to avoiding or reducing any liability to tax.” [6] Satisfaction of either condition is sufficient to negate the disallowance.

This principle means that even where genuine beneficial ownership has shifted beyond the 49% threshold, a taxpayer who can establish to the Assessing Officer that the restructuring was not tax-motivated can still preserve its accumulated losses. This second limb is, however, a factual determination susceptible to subjective assessment by Revenue officers, and in practice the burden of proof that the restructuring had no tax-avoidance motive rests heavily on the taxpayer.

The Mumbai ITAT: When Section 79 Is Triggered

The Mumbai Bench of the Income Tax Appellate Tribunal brought procedural clarity in Sodexo India Services Pvt. Ltd. [7], holding that Section 79 is attracted only in the year in which set-off is actually claimed, not in the year when the change in shareholding occurs. The Tribunal also reaffirmed that where the ultimate beneficial ownership of the company has remained unchanged, Section 79 cannot be invoked merely because the registered holder has changed. Since Sodexo India’s shares had moved between entities within the same group while the ultimate parent remained constant, the revisionary proceedings invoking Section 79 were held unsustainable. [7]

The timing clarification carries a double-edged consequence. It means that a restructuring that looks clean when executed could be reviewed under the shareholding composition that exists at the time set-off is eventually claimed, potentially years later. Careful monitoring of shareholder registers over the entire loss utilisation period is therefore essential for any closely held company with accumulated losses.

Where the Framework Falls Short

Despite the 2017 carve-out, Section 79 remains structurally inadequate for the startup funding lifecycle. The “all shareholders” condition is commercially unrealistic — early-stage investors, employee option-holders who exercise and sell, and seed-round angels routinely exit before a startup reaches profitability. The seven-year limitation compounds this: startups in capital-intensive industries — deep technology, life sciences, hardware, electric vehicles — commonly do not reach profitability within seven years, and once losses fall outside the seven-year window the startup carve-out ceases to apply. [8] The turnover threshold under Section 80-IAC adds a further complication, since a company that crosses the relevant threshold in a later year may find that it no longer qualifies as an eligible startup when it attempts to claim set-off, even though it unquestionably qualified when the losses were incurred.

For startups receiving foreign venture capital through offshore holding structures — near-universal in Indian tech startups — the Delhi High Court’s insistence on explicit documentary evidence of beneficial ownership creates structural risk. A fund manager sitting in Singapore who invested through a Mauritius SPV is the beneficial owner in any economic sense, but the Revenue may treat the SPV as the beneficial shareholder for Section 79 purposes. The PwC analysis of the Mumbai ITAT’s 2021 ruling, which denied Section 79’s application where voting power and beneficial ownership effectively remained unchanged after an intra-group merger, shows that taxpayers can succeed on these facts — but the litigation cost and uncertainty remain significant. [9]

Conclusion

Section 79 of the Income Tax Act, 1961 was designed to prevent tax trafficking in losses. Over six decades of litigation it has evolved into something considerably more complex — a provision that routinely catches legitimate group reorganisations and startup funding rounds within its net, while courts across the country disagree about whether and how beneficial ownership arguments can displace its application. The 2017 startup carve-out was well-intentioned but too narrowly drawn: the “all shareholders” condition is commercially impractical, the seven-year window is insufficient for many industries, and DPIIT recognition introduces administrative fragility into what ought to be a tax-stable relationship. Until Parliament revisits Section 79 with genuine attention to the startup funding lifecycle, or until the Supreme Court resolves the conflict between the Karnataka and Delhi High Courts on beneficial ownership, taxpayers will continue to face avoidable loss forfeitures that have nothing to do with tax avoidance.

References

[1] Section 79, Income Tax Act, 1961, as amended by Finance (No. 2) Act, 2019 — https://indiankanoon.org/doc/814605/

[2] PwC India, “Section 79 and its implications on global and domestic transactions/restructuring,” Tax Guru, June 2022 — https://taxguru.in/income-tax/section-79-implications-global-domestic-transactions-restructuring.html

[3] Argus Partners, “Extended Set Off and Carry Forward Period under Section 79 of the IT Act” — https://www.argus-p.com/updates/updates/extended-set-off-and-carry-forward-period-under-section-79-of-the-it-act/

[4] CIT v. AMCO Power Systems Ltd., Karnataka High Court, ITA Nos. 766, 765, 767, 769 of 2009 and 1046 of 2008 — https://itatonline.org/archives/cit-vs-amco-power-systems-ltd-karnataka-high-court-s-79-as-the-purpose-of-the-provision-is-to-prevent-misuse-of-losses-by-transferring-ownership-it-should-be-restricted-to-cases-of-transfer-of-be/

[5] Yum Restaurants (India) Pvt. Ltd. v. ITO, Delhi High Court, ITA Nos. 349 and 388 of 2015, decided 13 January 2016 — https://indiankanoon.org/doc/85916376/

[6] CIT, Bombay v. Italindia Cotton Co. (P) Ltd., Supreme Court of India, Civil Appeal No. 1520(NT) of 1986, decided 5 September 1988 — https://courtverdict.com/supreme-court-of-india/the-commissioner-of-income-tax-bombay-vs-ms-italindia-cotton-co-p-ltd

[7] S.R. Patnaik, “Section 79 cannot be invoked when there is no change in ultimate beneficial shareholding,” Cyril Amarchand Blogs, March 2023 — https://tax.cyrilamarchandblogs.com/2023/03/section-79-cannot-be-invoked-when-there-is-no-change-in-ultimate-beneficial-shareholding/

[8] “Section 79: Carry Forward and Set Off of Losses in Case of Eligible Startups,” Tax Guru, August 2019 — https://taxguru.in/income-tax/section-79-carry-set-loss-case-ofeligible-startups-condition-relaxed.html

[9] PwC India, “Applicability of Section 79 of the Act denied where there is no change in voting power and beneficial ownership,” Tax Insights, September 2021 — https://www.pwc.in/assets/pdfs/news-alert/tax-insights/2021/pwc_tax_insights_15_september_2021_applicability_of_section_79_of_the_act_denied_where_there_is_no_change_in_voting_power_and_beneficial_ownership.pdf

The post Startup Losses and Section 79 of the Income Tax Act, 1961: When Anti-Abuse Rules Kill Legitimate Restructuring appeared first on Bhatt & Joshi Associates.

The proliferation of artificial intelligence-generated synthetic media has created unprecedented challenges for digital governance worldwide. In India, the Ministry of Electronics and Information Technology notified the Information Technology (Intermediary Guidelines and Digital Media Ethics Code) Amendment Rules, 2026 (IT Amendment Rules 2026) on February 10, 2026, which came into force on February 20, 2026 [1]. These amendments introduce stringent timelines for content takedown, particularly a two-hour window for removing non-consensual intimate images and deepfake pornography, raising critical questions about constitutional validity and practical feasibility. This article examines whether the IT Amendment Rules 2026 strike a proportionate balance between protecting fundamental rights and ensuring operational viability for digital intermediaries.

The Regulatory Framework: Understanding the IT Amendment Rules 2026

The Information Technology Act, 2000 serves as the foundational legislation governing cyberspace in India, with the IT Rules 2021 providing detailed guidelines for intermediary liability. The IT Amendment Rules 2026 specifically target synthetically generated information, defined under the newly inserted Rule 2(1)(wa) as “audio, visual or audio-visual information which is artificially or algorithmically created, generated, modified or altered using a computer resource, in a manner that such information appears to be real” [1].

Under the amended framework, intermediaries must now remove content within drastically compressed timelines. Rule 3(1)(d) mandates removal of unlawful content within three hours of receiving a government or court order, reduced from the previous thirty-six hour window [2]. More significantly, Rule 3(2)(b) requires intermediaries to act within two hours for cases involving exposure of private areas, nudity, sexual acts, or artificially morphed images that were previously subject to a twenty-four hour deadline [3].

The amendments also impose mandatory labelling requirements under Rule 4(1A), requiring significant social media intermediaries to ensure users declare whether uploaded content is synthetically generated, and to embed permanent metadata or unique digital identifiers in such content [2]. These provisions are designed to address the exponential rise in deepfake-related crimes, with Indians losing approximately twenty-two thousand eight hundred forty-five crore rupees to cybercriminals in 2024, marking a two hundred six percent increase from the previous year [1].

Constitutional Foundations: Article 19 and the Freedom of Speech Framework

The constitutional validity of content takedown regulations must be examined through the lens of Article 19(1)(a) of the Constitution of India, which guarantees all citizens the right to freedom of speech and expression. This right extends to digital platforms and online speech, as established in numerous Supreme Court pronouncements. However, Article 19(2) permits the state to impose reasonable restrictions on this freedom in the interests of sovereignty and integrity of India, security of the state, friendly relations with foreign states, public order, decency or morality, contempt of court, defamation, or incitement to an offense.

The landmark judgment in Shreya Singhal v. Union of India [4] fundamentally reshaped intermediary liability law in India. The Supreme Court struck down Section 66A of the Information Technology Act, 2000 for being unconstitutionally vague and having a chilling effect on free speech. More importantly for the present discussion, the Court read down Section 79 of the IT Act and Rule 3(4) of the Intermediaries Guidelines to mean that intermediaries obtain actual knowledge requiring content takedown only through a court order or notification from a government authority, not through private complaints.

Justice Nariman observed in Shreya Singhal that “adjudicating on whether or not there is contravention of a particular provision of law, is the quintessential sovereign function to be discharged by the State or its organs. This function cannot be delegated to private parties such as intermediaries” [4]. This principle remains foundational to understanding the scope and limits of intermediary obligations under Indian law.

The tension between free expression and content moderation has been further explored in recent jurisprudence. The Constitution bench in recent observations emphasized that restrictions on speech must be precisely tailored, proportionate, and narrowly drawn to pass constitutional scrutiny. Any framework limiting expression must not be ambiguous or overbroad, and must serve a legitimate state interest through the least restrictive means available.

Proportionality Analysis: Balancing Rights and Regulatory Objectives

The proportionality test, derived from constitutional jurisprudence, requires that any restriction on fundamental rights must satisfy four criteria: it must have a legitimate aim, be suitable to achieve that aim, be necessary in that no less restrictive alternative exists, and maintain a fair balance between the restriction and the rights affected.

The legitimate aim of the two-hour takedown window is clear and compelling. Non-consensual intimate imagery and deepfake pornography cause severe psychological trauma, reputational damage, and constitute violations of dignity and privacy. These harms are often irreversible, with content spreading rapidly across platforms and causing lasting damage to victims. The Supreme Court of India, including the Chief Justice himself who became a victim of a deepfake video, has repeatedly flagged the inadequacy of existing laws in addressing this digital menace [5].

However, the necessity prong of the proportionality test raises significant concerns. A two-hour response window for global platforms handling millions of content pieces daily presents formidable operational challenges. Automated detection systems, while increasingly sophisticated, struggle with accuracy rates and generate both false positives and false negatives. Human moderation at scale within such compressed timelines requires substantial infrastructure investment, multilingual expertise, and contextual understanding that may not be immediately available.

Furthermore, the rules do not provide clear standards for what constitutes “reasonable and appropriate technical measures” for detecting prohibited synthetic content, nor do they establish performance benchmarks or acceptable error-rate thresholds [3]. This ambiguity creates uncertainty for intermediaries attempting compliance while simultaneously risking over-censorship to avoid liability.

The Deepfake Crisis: Judicial Recognition and Response

Indian courts have increasingly recognized the unique threats posed by deepfake technology. In Arun Jaitley v. Network Solutions Private Limited, the Delhi High Court protected personality rights in the digital domain, establishing that personal names of prominent individuals merit protection against cybersquatting and unauthorized use [6]. While this case predated the deepfake era, its reasoning about protecting digital identity and preventing misuse of persona has been extended to contemporary challenges.

More recently, courts have addressed deepfake-specific harms. The Delhi High Court, in addressing cases involving prominent personalities, has issued orders requiring platforms to deploy automated technology for detecting and deleting infringing content. These judicial directions acknowledge that manual takedown procedures are inadequate for addressing the scope and velocity of digital harm, necessitating technological solutions to counter technological threats [7].

The Supreme Court’s jurisprudence on dignity and privacy rights under Article 21 provides additional constitutional grounding for robust anti-deepfake measures. The right to life and personal liberty has been interpreted expansively to include the right to dignity, privacy, and reputation. Non-consensual intimate imagery, whether real or synthetic, violates these fundamental rights in ways that justify state intervention.

Comparative Perspectives: Global Approaches to Deepfake Regulation

India’s two-hour takedown mandate can be contextualized against international regulatory approaches. The United States enacted the Take It Down Act in May 2025, requiring platforms to remove non-consensual intimate imagery and deepfakes within forty-eight hours of notification [8]. This legislation provides more time for compliance while establishing federal standards for notice-and-takedown procedures.

The European Union’s approach under the AI Act and the Digital Services Act establishes risk-based frameworks that impose heightened obligations on very large online platforms while providing more nuanced timelines and procedural safeguards. These frameworks recognize that different types of content and different platform capacities warrant differentiated regulatory responses.

The critical distinction in India’s approach is the extremely compressed timeline coupled with potential loss of safe harbor immunity under Section 79 of the IT Act for non-compliance. This creates high-stakes pressure on intermediaries that may incentivize over-removal of content to avoid liability, potentially infringing on legitimate speech and expression.

Operational Feasibility: The Implementation Challenge

The operational challenges of implementing a two-hour takedown window cannot be understated. Platforms must first receive notification, verify the complainant’s identity and claim, locate the specific content across potentially multiple instances and formats, assess whether it genuinely violates the rules rather than constituting legitimate parody or satire, and then execute technical removal while maintaining records for potential legal challenges.

For global platforms operating across time zones, the requirement means maintaining round-the-clock moderation teams with expertise in Indian law and cultural context. For smaller intermediaries and emerging platforms, these requirements may create insurmountable barriers to entry, potentially consolidating the digital marketplace in favor of large incumbents with resources to build extensive compliance infrastructure.

The IT Amendment Rules 2026 provide limited clarity on contentious edge cases. Exclusions for “routine editing” and “good faith creation” remain subject to interpretation, particularly for satire, parody, or artistic expression [3]. The mechanism for verifying user declarations about synthetic content is also undefined, leaving intermediaries to develop their own standards without regulatory guidance.

Furthermore, the rules do not address the reality that deepfakes are constantly evolving technologically. Detection methods that work today may be obsolete tomorrow as generation techniques become more sophisticated. This creates an arms race dynamic where compliance frameworks must continuously adapt, yet the regulatory timelines remain fixed.

The Safe Harbor Dilemma: Balancing Protection and Accountability

Section 79 of the Information Technology Act provides intermediaries with safe harbor immunity from liability for third-party content, provided they comply with due diligence requirements. The Shreya Singhal judgment clarified that this immunity is preserved when intermediaries respond appropriately to government or court orders for content takedown [4].

The IT Amendment Rules 2026 explicitly state that intermediaries will not lose safe harbor protection when removing or disabling access to synthetically generated content in accordance with the rules [2]. However, the practical effect of the compressed timelines is to shift substantial risk to intermediaries. Failure to remove content within two hours could result in loss of immunity, exposing platforms to liability for damages suffered by victims.

This creates a strong incentive structure favoring over-removal. When faced with uncertainty about whether specific content violates the rules, platforms will likely err on the side of taking down questionable material rather than risking significant legal exposure. This dynamic undermines the careful balance struck in Shreya Singhal, where the Court sought to prevent intermediaries from becoming private judges of content legality.

The constitutional concern is that this effectively delegates quasi-judicial functions to private platforms, requiring them to make rapid determinations about content legality without the procedural safeguards that accompany governmental or judicial decision-making. This runs contrary to the Shreya Singhal principle that adjudicating legal violations is a quintessentially sovereign function.

Recommendations: Toward a More Balanced Framework

A more constitutionally sound and operationally viable framework would incorporate several modifications. First, the rules should establish clear, graduated timelines based on content type and harm severity. Genuinely harmful non-consensual intimate imagery might warrant expedited removal, while other synthetic content could operate under longer timeframes allowing for careful review.

Second, procedural safeguards must be strengthened. Users whose content is removed should receive notification and have meaningful opportunity for appeal. The rules should establish independent review mechanisms, similar to content review boards that some platforms have voluntarily adopted, ensuring that takedown decisions are subject to oversight beyond the initial platform determination.

Third, the regulatory framework should provide clearer technical standards and guidance. Rather than leaving intermediaries to develop their own detection methodologies, the government could establish certification programs for detection tools, create safe harbors for good faith use of approved technologies, and provide regular guidance on emerging deepfake techniques and appropriate responses.

Fourth, the rules should explicitly protect legitimate uses of synthetic media. Clear carve-outs for news reporting, academic research, artistic expression, and political commentary would prevent over-censorship while still addressing genuinely harmful content. These exceptions should be defined with sufficient precision to provide meaningful guidance while remaining flexible enough to accommodate technological evolution.

Finally, enforcement should be proportionate and consider platform size and resources. Differential standards for large social media intermediaries versus smaller platforms would recognize that compliance capacity varies substantially across the digital ecosystem. This tiered approach is common in other jurisdictions and helps prevent regulatory capture by large incumbents.

Conclusion

The Information Technology (Intermediary Guidelines and Digital Media Ethics Code) Amendment Rules, 2026 represent India’s most comprehensive effort to address the deepfake crisis. The two-hour takedown window for non-consensual intimate imagery reflects legitimate concerns about severe harms that victims suffer from such content. However, the constitutional validity and operational feasibility of this extremely compressed timeline remain questionable.

The framework must be evaluated against the standards established in Shreya Singhal v. Union of India and the broader constitutional jurisprudence on freedom of speech and expression. While protecting victims of deepfake abuse is a compelling state interest, the means chosen must be narrowly tailored, provide adequate procedural safeguards, and avoid creating incentive structures that lead to over-censorship of legitimate speech.

The tension between rapid response to digital harm and protection of free expression is not unique to India, but India’s approach is among the most aggressive globally. As implementation proceeds, close monitoring of compliance rates, false positive removals, and impact on legitimate speech will be essential. The rules include provisions for periodic review, and such reviews should incorporate empirical data on implementation challenges and constitutional concerns.

Ultimately, effective deepfake regulation requires a multi-stakeholder approach combining legal frameworks, technological solutions, media literacy, and international cooperation. The two-hour takedown window, while well-intentioned, may prove to be operationally impossible without substantial modifications that better balance the legitimate interests of all stakeholders while maintaining fidelity to constitutional principles of free expression and due process.

References

[1] Ministry of Electronics and Information Technology. (2026). Information Technology (Intermediary Guidelines and Digital Media Ethics Code) Amendment Rules, 2026. https://www.outlookbusiness.com/news/meity-notifies-it-rules-to-curb-deepfakes-and-ai-generated-content

[2] Outlook Business. (2026). AI Labelling, Quicker Takedowns: Decoding India’s New Social Media Rules. https://www.outlookbusiness.com/explainers/ai-labelling-quicker-takedowns-decoding-indias-new-social-media-rules

[3] Obhan & Associates. (2026). India’s New Deepfake Regulation: MeitY Notifies Amendments to Information Technology Rules 2021. https://www.obhanandassociates.com/blog/indias-new-deepfake-regulation-meity-notifies-amendments-to-information-technology-rules-2021/

[4] Shreya Singhal v. Union of India, (2015) 5 SCC 1, AIR 2015 SC 1523. https://indiankanoon.org/doc/110813550/

[5] The Sentinel Assam. (2026). Can new IT rules stop the deepfake epidemic? https://www.sentinelassam.com/more-news/editorial/can-new-it-rules-stop-the-deepfake-epidemic

[6] Arun Jaitley v. Network Solutions Private Limited, CS(OS) 1745/2009, Delhi High Court (2011). https://indiankanoon.org/doc/754672/

[7] Khurana & Khurana. (2025). Deepfake Regulation India 2025: MeitY’s Comprehensive IT Rules Amendment. https://www.khuranaandkhurana.com/deepfake-regulation-india-2025-meity-s-comprehensive-it-rules-amendment

[8] Skadden, Arps, Slate, Meagher & Flom LLP. (2025). ‘Take It Down Act’ Requires Online Platforms To Remove Unauthorized Intimate Images and Deepfakes When Notified. https://www.skadden.com/insights/publications/2025/06/take-it-down-act

[9] The Federal. (2026). India mandates 3-hour takedown for AI content: FAQ of what you need to know. https://thefederal.com/category/explainers-2/ai-content-faq-on-new-it-rules-for-ai-generated-content-deepfake-229394

The post MeitY’s 2-Hour Deepfake Takedown Window Under IT Amendment Rules 2026: Constitutionally Proportionate or Operationally Impossible? appeared first on Bhatt & Joshi Associates.

The question of whether an artificial intelligence platform can qualify as an “intermediary” under Indian law — and thereby claim the protection of safe harbour under Section 79 of the Information Technology Act, 2000 — is one of the most pressing and underexamined questions in Indian technology law today. For more than two decades, Section 79 has functioned as the backbone of India’s internet economy, shielding platforms from secondary liability for third-party content. The provision was drafted at a time when the internet was imagined as a passive pipe: a conduit through which users sent and received information. Algorithms of the generative and recommending kind that now define digital experience were simply not contemplated [1].

Today, platforms such as YouTube, Instagram, and AI-native services like Grok do not simply host content. Their algorithms curate, amplify, personalise, and in the case of generative AI, actively produce it. This makes the question far from academic: if an algorithm is found to be an active participant in content creation or curation, the platform deploying it may lose its statutory shield entirely. The Ministry of Electronics and Information Technology (MeitY) has, through a series of advisories in 2023 and 2024, begun to signal precisely this shift — that AI is not simply content hosted on a platform, but content shaped and generated by it [2].

The Architecture of Section 79 of the IT Act: What the Provision Actually Says

Section 79 of the Information Technology Act, 2000, provides in its operative part: “Notwithstanding anything contained in any law for the time being in force but subject to the provisions of sub-sections (2) and (3), an intermediary shall not be liable for any third party information, data, or communication link made available or hosted by him.” This immunity is not unconditional. Sub-section (2) requires that the intermediary must not have initiated the transmission, must not have selected the receiver, and must not have selected or modified the information contained in the transmission. It must also observe due diligence and comply with the guidelines prescribed by the Central Government.

Sub-section (3) withdraws the protection in two scenarios: first, where the intermediary has conspired with, abetted, aided, or induced the commission of an unlawful act; and second, where the intermediary, upon receiving “actual knowledge” that unlawful content is being hosted on its platform, fails to expeditiously remove or disable access to that material. The term “intermediary” is defined under Section 2(1)(w) of the IT Act as “any person who on behalf of another person receives, stores or transmits that record or provides any service with respect to that record,” and expressly includes telecom service providers, internet service providers, web-hosting service providers, search engines, online payment sites, online-auction sites, online marketplaces, and cyber cafes [1].

The structure of this provision assumes a fundamental premise: that the intermediary is a passive actor. Its immunity is premised on its not having shaped the content in question. The moment it crosses into active participation — selecting, modifying, inducing — the statutory protection falls away. The rise of AI platforms tests every element of this assumption.

Shreya Singhal v. Union of India (2015): The Constitutional Baseline

No discussion of Section 79 of the IT Act is complete without a reckoning with the Supreme Court’s landmark judgment in Shreya Singhal v. Union of India, (2015) 5 SCC 1, delivered on 24 March 2015 by a bench of Justices J. Chelameswar and R.F. Nariman. The case arose from a batch of writ petitions under Article 32 of the Constitution of India, principally challenging the constitutionality of Sections 66A, 69A, and 79 of the IT Act. The Supreme Court’s treatment of Section 79 fundamentally reshaped the intermediary liability regime in India [3].

The Court read down Section 79(3)(b) to narrow its scope significantly. The holding was unambiguous:

“Section 79 is valid subject to Section 79(3)(b) being read down to mean that an intermediary upon receiving actual knowledge from a court order or on being notified by the appropriate Government or its agency that unlawful acts relatable to Article 19(2) are going to be committed then fails to expeditiously remove or disable access to such material.”

In practical terms, the Court held that intermediaries are not required to act upon private takedown requests. “Actual knowledge,” as used in Section 79(3)(b), was interpreted to mean knowledge received through the medium of a court order — not a complaint from a private party. This interpretation rested on a practical foundation: holding intermediaries like Google and Facebook to a standard of responding to every private complaint would make it impossible for them to function, since millions of requests are received and an intermediary cannot be expected to adjudicate the legality of each piece of content on its own. The Court further affirmed that there is no positive obligation on intermediaries to monitor content on their platforms [3]. This no-monitoring principle remains foundational to India’s safe harbour regime under Section 79 of the IT Act, even as AI regulation begins to chip away at it.

Active vs. Passive Intermediaries: The Christian Louboutin Standard

The passive/active distinction now central to the AI liability debate was crystallised in Indian jurisprudence by the Delhi High Court in Christian Louboutin SAS v. Nakul Bajaj & Ors., 2018 SCC OnLine Del 12215, decided on 2 November 2018 by Justice Prathiba M. Singh. The case involved the luxury shoe brand’s claim against darveys.com, an e-commerce platform that used the plaintiff’s trademarks as meta-tags and claimed to sell authentic goods sourced from authorised stores [4].

The defendant’s principal defence was that it was a mere intermediary under Section 79 of the IT Act. Justice Singh rejected this defence and, in doing so, laid down a twenty-six point framework to determine whether an online platform is a passive conduit or an active participant. The court reasoned that so long as a platform acts as “mere conduit or passive transmitters of the records or of the information, they continue to be intermediaries, but merely calling themselves as intermediaries does not qualify all e-commerce platforms or online market places as one.” The court then held:

“When an e-commerce website is involved in or conducts its business in such a manner, which would see the presence of a large number of elements enumerated above, it could be said to cross the line from being an intermediary to an active participant.”

By curating product listings, arranging logistics, using meta-tags, and guaranteeing authenticity, darveys.com had exceeded the role of a neutral conduit. The court also held that failure to observe due diligence with respect to intellectual property rights could amount to “conspiring, aiding, abetting, or inducing” unlawful conduct under Section 79(3)(a), independently disentitling the platform from safe harbour [4].

This framework applies with full force to AI platforms. When a recommendation algorithm selects which content a user sees, or when a generative AI model produces text or video in response to a user prompt, the question of whether these functions constitute “selection” or “modification” of information within the language of Section 79(2)(b) becomes the defining legal inquiry. The Christian Louboutin standard supplies the doctrinal tool; generative AI supplies the stress test.

IT (Intermediary Guidelines and Digital Media Ethics Code) Rules, 2021: Expanding the Compliance Perimeter

The Information Technology (Intermediary Guidelines and Digital Media Ethics Code) Rules, 2021, notified on 25 February 2021 under Section 87 read with Section 79 of the IT Act, represent the most significant regulatory expansion of intermediary obligations since the original 2011 Guidelines. Rule 7 makes explicit that an intermediary which fails to comply with prescribed due diligence requirements shall no longer be entitled to safe harbour under Section 79(1) of the IT Act and shall be liable under applicable laws [1].

The 2021 Rules introduced the classification of “significant social media intermediaries” (SSMIs) — social media intermediaries with more than fifty lakh (five million) registered users in India. SSMIs bear substantially heavier obligations: they must appoint a Chief Compliance Officer, a Grievance Redressal Officer, and a Nodal Contact Person, all resident in India. Rule 4(2) requires SSMIs that primarily provide messaging services to enable identification of the “first originator” of information where directed by a court or competent authority under Section 69 of the IT Act.

For AI platforms, the most consequential provision is Rule 3(1)(b), which requires intermediaries to “make reasonable efforts by itself, and to cause the users of its computer resource” not to publish certain categories of prohibited content. This language has been interpreted as potentially imposing a preventive obligation — not merely reactive removal — that moves the compliance standard toward something approaching a monitoring duty. If AI systems deployed on a platform generate or amplify prohibited content, the question of whether the platform made “reasonable efforts” to prevent this, independently of any user action, becomes immediately live [2].

MeitY’s AI Advisories: The Regulatory Turn

India’s formal attempt to address AI within the intermediary liability framework began in November 2023 and crystallised through MeitY advisories issued in early 2024. The March 15, 2024 Advisory — which replaced the March 1, 2024 Advisory — directed intermediaries to ensure that the use of “AI models, large language models, generative AI technology, software or algorithms” on or through their platforms does not allow users to host, display, upload, modify, publish, transmit, store, update, or share any content in violation of the Intermediary Guidelines or any other law in force [2].

The advisory’s significance lies in its implicit treatment of AI not as content but as a potentially liable actor within the intermediary ecosystem. By requiring platforms to ensure that AI models deployed on them do not enable unlawful conduct, MeitY effectively placed the responsibility for AI-generated harm squarely on the platform. A platform that deploys a generative AI model which produces deepfake content, defamatory material, or content that undermines democratic processes cannot credibly claim it was merely hosting third-party information — because the AI is not a third party in any conventional sense. It is the platform’s own deployed technology [2].

The advisories also addressed deepfakes specifically, reflecting the 2023 Rashmika Mandanna incident, where AI-generated synthetic video caused significant public and political concern. That episode illustrated how AI-generated content can cause reputational harm at a scale and speed that outpaces any traditional notice-and-takedown mechanism, and demonstrated to MeitY that the existing framework needed explicit AI-specific obligations [5].

IT (Intermediary Guidelines) Amendment Rules, 2026: Formalising AI Liability

The most direct regulatory intervention to date is the Information Technology (Intermediary Guidelines and Digital Media Ethics Code) Amendment Rules, 2026, notified by MeitY on 20 February 2026. These rules, for the first time, introduce a statutory definition of “synthetically generated information” (SGI), described as any content that is artificially or algorithmically created, generated, modified, or altered using a computer resource in a manner that appears authentic. This definition is intentionally broad, capturing the full range of AI-generated content including deepfakes, synthetic audio-visual material, and algorithmically altered images [5].

The 2026 Rules impose mandatory labelling obligations on intermediaries that facilitate the creation of SGI. Visual content must carry a clear and permanent metadata identifier covering at least ten percent of the display area; audio content must contain an audible disclosure during at least ten percent of its duration. These labels cannot be removed, modified, or suppressed by users. The rules also dramatically reduce takedown timelines: unlawful or prohibited AI-generated content must be removed or disabled within three hours of receiving a lawful notice [5].

The 2026 Rules expressly clarify that intermediaries acting in good faith and in compliance with these obligations will continue to enjoy safe harbour protection under Section 79 of the IT Act. Conversely, failure to comply — failure to label, delay in takedown, or inadequate grievance handling — may result in the loss of that protection. Safe harbour is thereby transformed from a passive shield into a compliance-contingent privilege. The standard is no longer merely reactive: an intermediary must demonstrate system-level preparedness to deal with AI-generated risks proactively, not merely respond to them after harm has occurred [5].

The Grok Question: When AI Is the Platform

The most pointed articulation of the AI-as-creator problem in Indian regulatory discourse concerns the deployment of Grok, an AI model integrated into X (formerly Twitter). The Indian government has argued — publicly, if not yet conclusively in litigation — that X’s deployment of Grok effectively makes it a creator of content, not merely a host. If Grok generates content in response to user prompts, X cannot claim to be a neutral intermediary whose only role is the passive transmission of third-party information. On this view, Section 79’s safe harbour would not apply, because the platform itself is the origin point of at least some of the content on it [6].

This is the active/passive distinction from Christian Louboutin transposed directly onto generative AI. The legal framework as it currently stands does not offer a clean answer. The definition of intermediary in Section 2(1)(w) refers to a person who “receives, stores or transmits” electronic records or “provides any service with respect to that record.” A generative AI model arguably does none of these things in the traditional sense — it creates records rather than receiving or transmitting them [1][6].

Researchers at the Carnegie Endowment have observed that existing definitions under the IT Act, when applied to AI systems, are “being stretched too thin” and that “generative AI systems may not fall neatly within the purview of either publisher or intermediary” under the current statutory framework [7]. This definitional gap is precisely why the 2026 Amendment Rules and the anticipated Digital India Act are significant: they represent attempts to fill a statutory vacuum that the original IT Act, drafted in 2000, could not have anticipated.

MySpace Inc. v. Super Cassettes Industries Ltd.: The No-Monitoring Principle and Its Limits

The no-monitoring principle affirmed in Shreya Singhal was reaffirmed by a Division Bench of the Delhi High Court in MySpace Inc. v. Super Cassettes Industries Ltd., (2017) 236 DLT 478. The court held that intermediaries are not under any positive obligation to proactively monitor content on their platforms for copyright infringement, and that “actual knowledge” must be in the form of a court order — not constructive or inferred knowledge. The court expressly rejected the argument that a platform’s technical ability to detect infringing content was equivalent to legal knowledge sufficient to impose liability [8].

This principle sits uneasily alongside the 2026 Rules’ mandatory labelling and three-hour takedown obligations for AI-generated content. If a platform deploys an AI model that generates content, and that content turns out to be unlawful, the platform’s argument that it had no “actual knowledge” of the specific unlawfulness is considerably weakened — because the AI is the platform’s own system. The content did not arrive from an unknown third-party originator; it was produced by the platform’s own technology. The no-monitoring principle was premised on the practical impossibility of reviewing every piece of user-generated content. That impossibility argument does not translate cleanly to AI-generated content, which the platform’s own systems produced and could, in principle, have been designed to screen from the outset [8].

X Corp. v. Union of India: Section 79(3)(b) and the Live Battleground of Safe Harbour

The question of how Section 79(3)(b) interacts with AI-generated content is being contested in live litigation before the Karnataka High Court in X Corp. v. Union of India, a writ petition filed on 5 March 2025 before Justice M. Nagaprasanna. X Corp. challenges the legality of information-blocking orders issued by various government ministries under Section 79(3)(b), following a MeitY Office Memorandum of 31 October 2023 that authorised all central ministries, state governments, and local police officers to issue content blocking orders through the Sahyog portal [9].

X’s core argument, drawing expressly on Shreya Singhal, is that Section 79(3)(b) cannot function as an independent mechanism for content blocking. Content blocking, X submits, can only occur through the constitutionally safeguarded process under Section 69A of the IT Act, which requires reasoned orders and procedural safeguards. By contrast, Section 79(3)(b) merely describes the circumstances in which safe harbour is lost — it does not independently confer blocking power on the executive [9]. For AI platforms, the implications are significant: if informal government notices under Section 79(3)(b) are sufficient to trigger takedown obligations for AI-generated content, platforms will face executive pressure to remove such content without judicial oversight, fundamentally altering the architecture of safe harbour from an immunity into a tool of executive content governance.

Conclusion

Section 79 of the IT Act was not written for the age of algorithms. Its passive-intermediary model, refined through case law from Shreya Singhal to Christian Louboutin to MySpace, assumes a clean separation between the platform and the content it hosts. Generative AI destroys that separation. When an algorithm recommends, curates, or creates content, the platform is no longer merely a conduit — it is a participant. Whether courts will treat that participation as sufficient to strip safe harbour protection depends on how the active/passive distinction is applied to algorithmic conduct. MeitY’s 2026 Amendment Rules have begun to answer this question legislatively, by conditioning safe harbour on demonstrated compliance with AI-specific obligations, mandatory labelling, and accelerated takedown timelines. The answer, in short, is that an algorithm can be treated as part of the intermediary for regulatory purposes — but the intermediary that deploys it cannot hide behind Section 79 when the algorithm itself is the source of the harm.

References

[1] Information Technology Act, 2000, Sections 2(1)(w) and 79, Ministry of Electronics and Information Technology, Government of India. Available at: https://www.indiacode.nic.in/show-data?actid=AC_CEN_45_76_00001_200021_1517807324077&orderno=105

[2] S&R Associates, “Investing in AI in India (Part 3): AI-related Advisories Under the Intermediary Guidelines,” October 2024. Available at: https://www.snrlaw.in/investing-in-ai-in-india-part-3-ai-related-advisories-under-the-intermediary-guidelines/

[3] Shreya Singhal v. Union of India, (2015) 5 SCC 1, Supreme Court of India, 24 March 2015. Full judgment available at: https://globalfreedomofexpression.columbia.edu/wp-content/uploads/2015/06/Shreya_Singhal_vs_U.O.I_on_24_March_2015.pdf

[4] Christian Louboutin SAS v. Nakul Bajaj & Ors., 2018 SCC OnLine Del 12215, Delhi High Court, 2 November 2018. Available at: https://indiankanoon.org/doc/99622088/

[5] TBA Law, “India’s IT Intermediary Rules 2026 Amendment on AI-Generated Content: A Legal Analysis,” 2026. Available at: https://www.tbalaw.in/post/india-s-it-intermediary-rules-2026-amendment-on-ai-generated-content-a-legal-analysis

[6] IAS Gyan, “Grok Case Raises Questions of AI Governance,” 2024. Available at: https://www.iasgyan.in/daily-editorials/grok-case-raises-questions-of-ai-governance

[7] Carnegie Endowment for International Peace, “India’s Advance on AI Regulation,” November 2024. Available at: https://carnegieendowment.org/research/2024/11/indias-advance-on-ai-regulation?lang=en

[8] Bar and Bench, “Generative AI and Intermediary Liability Under the Information Technology Act” (discussing MySpace Inc. v. Super Cassettes Industries Ltd., (2017) 236 DLT 478). Available at: https://www.barandbench.com/view-point/generative-ai-and-intermediary-liability-under-the-information-technology-act

[9] SC Observer, “X Relies on ‘Shreya Singhal’ in Arbitrary Content-Blocking Case in Karnataka HC,” July 2025. Available at: https://www.scobserver.in/journal/x-relies-on-shreya-singhal-in-arbitrary-content-blocking-case-in-karnataka-hc/

The post Section 79 Safe Harbour and AI Platforms: Can an Algorithm Be an Intermediary Under Indian Law? appeared first on Bhatt & Joshi Associates.