Three Hours to Disappear: India's Second-Amendment IT Rules and the Ideology Filter on a Global South Platform
On Saturday 11 April 2026, six Indian press bodies — the Editors' Guild of India, DIGIPUB, the Press Club of India, the Indian Women's Press Corps, the Network of Women in Media India, and the Delhi Union of Journalists — gathered at the Press Club on Raisina Road and demanded the "unconditional withdrawal" of the Ministry of Electronics and Information Technology's Second Amendment to the IT Rules 2021. Three days later the Committee to Protect Journalists called the draft "a direct attack on independent journalists using digital platforms." Three days after that, Amnesty International India urged immediate withdrawal on free-speech and privacy grounds. The deadline to submit comments is 29 April. Running underneath is a rule already in force: since 20 February 2026, any Indian social-media intermediary has three hours to act on a government takedown order for "synthetically generated information," down from thirty-six.
The immediate story is two overlapping MeitY proposals — one notified, one in draft — that together rewire the legal architecture of online speech for 830 million Indian internet users. The structural story is that "deepfakes," "misinformation," and "national security" are doing the same work in 2026 that anti-communism rhetoric did in earlier media theory: pre-empting a proportionality analysis. Ownership, advertising, sourcing and flak explain why Meta and X and Google comply; the ideological framing — "deepfakes," "national security" — explains why the policy reads, in Apar Gupta's phrase to France 24, as "much more a design of political control than of constitutional objectives." The draft is what platform regulation looks like when the regulator is confident the platform cannot say no.
What happened this month
On 10 February 2026 MeitY notified the "Synthetically Generated Information" (SGI) rules, which took effect 20 February. They require every intermediary enabling AI-generated content to label it with provenance metadata, deploy automated detection, and act on takedown notices within three hours, down from thirty-six; Grievance Appellate Committee orders collapse from twenty-four hours to two. LiveLaw quoted industry executives calling it "the shortest content-takedown timeline prescribed by any government in the world." Missing the deadline costs the intermediary its Section 79 safe-harbour protection — a cliff, not a fine.
On 30 March MeitY published the far broader Draft IT Second Amendment Rules 2026, comments open to 29 April. It extends the Ministry of Information and Broadcasting's Digital Media Ethics Code — until now applied only to online news publishers and OTT platforms — to individual social-media users posting "news and current affairs content" (The Wire, 31 March). It amends Rule 8 to apply Rules 14 (Inter-Departmental Committee review), 15 (blocking directions) and 16 (emergency blocking without hearing) to those users and their platforms. And it conditions Section 79 safe harbour on "compliance with Ministry-issued advisories, directions and guidelines" — binding administrative instruments issued without legislative process.
Between the two moves, the orders kept coming. MediaNama's tracker logged more than forty government-ordered removals in March alone, including Facebook pages for National Dastak, Molitics and 4PM News, and orders against journalist Mohammed Zubair and satirist Rajeev Nigam. On 31 March the US Trade Representative's 2026 National Trade Estimate called India's takedown orders to American social-media firms "politically motivated" non-tariff trade barriers for a second consecutive year. On 11 April the Press Club convened; on 14 April CPJ; on 17 April Amnesty. That is the seven-day window.
What the primary documents actually say
The SGI notification borrows the EU AI Act's Article 50 transparency language almost verbatim, but yokes it to a three-hour removal clock Brussels does not impose. The Internet Freedom Foundation's problem is not the labelling — defensible — but the clock. "These impossibly short timelines," IFF wrote on 10 February, "eliminate any meaningful human review, forcing platforms toward automated over-removal and creating a prior-restraint regime incompatible with Article 19(1)(a)." That collides directly with the Supreme Court's 2015 judgment in Shreya Singhal v. Union of India, which read "actual knowledge" under Section 79 to mean a court order or formal government notification — not an algorithm's classification, not a three-hour reflex.
The March draft pretends to be a technical fix. A new proviso to Rule 8(1) extends the code of ethics governing online news publishers to "intermediaries insofar as they host, cause to be hosted, or make available user-generated content relating to news and current affairs on their platforms, and to users who publish such content." In plain English: the Ministry of Information and Broadcasting's Inter-Departmental Committee, a body of bureaucrats convened under Rule 14, can order the removal of a YouTube video, a Facebook post or an X thread by a freelance journalist or satirist, and the carrying platform faces loss of Section 79 protection if it resists. Prateek Waghre, fellow at Tech Policy Press and former IFF executive director, told France 24 the scope is "wider than anything before" and that "there isn't enough time for platforms to figure out the legality of a post. They're likely to just remove content without thinking about it." The Editors' Guild's 11 April resolution caught the mechanism in a phrase: the amendments build "an infrastructure of censorship" that makes independent digital journalism "financially terminal."
The structural frame: ideology on the stack
"Deepfakes," "synthetically generated information," "misinformation" and "national security" do in 2026 what anti-communism rhetoric did in earlier decades of media theory: they pre-empt a proportionality analysis. Nobody is for deepfakes. Once the frame is invoked, the discussion moves from "is this rule proportionate?" to "are you against the harm, yes or no?"
Intermediary liability rules are where state ideology gets converted into platform policy without democratic authorship; every compressed takedown window is an un-legislated speech rule shipped inside a compliance deadline. The automation-bias point follows: once platforms are pushed to automated detection by deadline, the bias is toward false positives, because a false positive costs the platform nothing and a false negative costs Section 79 immunity. Trust in platforms collapsed first in the Global North after Cambridge Analytica; in the Global South it collapsed after COVID misinformation, the 2024 Lok Sabha elections, and cheap vernacular-language deepfakes. A decade ago, shrinking a takedown window by eleven-twelfths would have required surviving a front-page editorial. In 2026 it takes a notification and a press release about "AI harms."
What the frame hides is the distributional question. Who loses three hours of due process? Not Meta, with its Bengaluru compliance team; not X, which has complied with Section 69A orders at what Al Jazeera called a "rubber-stamp" rate. The losers are the independent Hindi-language YouTube channels — Molitics, National Dastak, 4PM News — whose removed Facebook pages triggered CPJ's 14 April statement; the regional-language satirists; the Kashmiri reporters whose accounts have been withheld under Section 69A without reasons. Users with in-house lawyers survive. Users with a phone survive as long as the classifier does not mis-fire.
Historical precedent: the Shreya Singhal line and the FCU detour
Two Indian precedents bracket the moment. In 2015 the Supreme Court struck down Section 66A in Shreya Singhal v. Union of India and read down Section 79 to insulate intermediaries absent a court order or formal government notification under Article 19(2). IFF's position is that the 30 March draft drives directly through that judgment by rebranding administrative advisories as binding directions; MeitY's position, reported in The Print on 31 March, is that "advisories" are not blocking orders in the Section 69A sense and therefore not subject to the Shreya Singhal standard. Civil society's reply: loss of safe harbour is functionally equivalent to a blocking order, and the test should travel with the effect, not the label.
The second precedent is the 2023 "Fact Check Unit" amendment, struck down in September 2024 by the Bombay High Court in Kunal Kamra v. Union of India. Justice Chandurkar's tie-breaking opinion, quoting Shreya Singhal, held the FCU had a "chilling effect on free speech" and was ultra vires Article 19(1)(a). The 30 March draft revives, in wider and more diffuse form, the architecture the FCU attempted: the state, acting through executive advisories, shapes what intermediaries carry, with loss of safe harbour as the coercive lever. The lesson the bureaucracy drew from Kunal Kamra was not that the mechanism is unconstitutional; it is that it needs to be distributed across more rules, framed through the ideology of "deepfakes" rather than "false or misleading," and shipped with a thirty-day comment window so narrow that civil society has to triage.
What it means for information control going forward
Three implications follow.
First, the intermediary liability carve-out that held the internet's shape since the United States' Section 230 and India's Shreya Singhal is now a live negotiation in every Global South jurisdiction. Brazil's PL 2630, Nigeria's NBC code, Indonesia's MR5 and South Africa's Films and Publications Board amendments all carry variants of the same move — conditioning safe harbour on compliance with executive directions, branded under the ideology of "misinformation" or "harmful content." What happens in the 29 April MeitY window sets a template much larger than India.
Second, the EU's parallel enforcement — the European Board for Digital Services' 18th meeting on 15 April endorsing the Commission's preliminary findings against TikTok, and the mid-April Article 40 researcher-access deadline for X — is not an opposite pole; it is the other end of the same arc. Brussels and New Delhi are both tightening intermediary obligations under a "protection" frame (minors in one, deepfakes in the other). The live question is whether the DSA's procedural architecture — risk assessments, auditors, researcher data access — travels southward, or whether the "three-hour clock" pattern travels north.
Third, the corporate-regulatory bargain has visibly flipped. In 2018 platforms still pushed back — the 2021 rules produced a WhatsApp constitutional challenge and a Twitter lawsuit in Karnataka. In 2026 Meta complies silently, Google complies silently, X complies selectively under Musk's own political calculus, and only Wikipedia has publicly litigated a content-related takedown. The platforms cannot say no: the cost of saying no is a Section 79 cliff; the cost of saying yes is absorbed into compliance overhead and priced into Indian-market CPMs. That is not a policy failure. That is the control mechanism operating as designed.
What Amnesty called "the chilling effect on free speech and privacy" is the sound of a mature filtration system. The surprising thing is not that India's executive wants it. The surprising thing is how little of the Western tech press has treated the 11 April resolution by India's own editors as the primary document it is.
That is the story.