Introduction:
Social media platforms have become one of the main public forums in India. They carry news, political debate, commercial speech and personal grievances at a scale no traditional publisher has ever handled. That reach has placed them at the centre of a question the Information Technology Act, 2000 was never designed to answer: when unlawful content appears online, how far should the platform that hosts it be held responsible?
Indian law has answered that question through the principle of safe harbour, which protects intermediaries from liability for third-party content so long as they observe due diligence and act on lawful orders. Over the past eighteen months, that framework has changed considerably. Amendments to the IT Rules, 2021 have cut takedown timelines to three hours, created obligations for AI-generated content and proposed tying immunity to compliance with executive advisories, while the courts are examining the fact-check unit and the Sahyog portal.
These are not only regulatory questions. Every duty placed on a platform shapes the speech and privacy of the people who use it, which brings Articles 14, 19(1)(a) and 21 of the Constitution squarely into play. The data protection regime that began taking effect in November 2025 adds a further layer of obligations, particularly for children’s data.
This article examines the statutory basis of intermediary liability in India, traces how that framework has been reshaped, and tests those changes against constitutional standards. It then compares the approaches of other jurisdictions and sets out what a balanced Indian framework could include.
The original bargain:
Section 79 of the Information Technology Act, 2000 gives intermediaries conditional immunity. A platform is not liable for third-party content it only hosts or transmits, so long as it does not initiate the transmission, choose the recipient or alter the content, and so long as it observes due diligence.
The condition that does the real work is Section 79(3)(b). A platform loses its immunity if, on receiving “actual knowledge” that content is unlawful, it fails to remove that content quickly.
In Shreya Singhal v. Union of India (2015), the Supreme Court read that phrase narrowly. Actual knowledge means a court order or a notification from the appropriate government, and only for content that falls within the grounds in Article 19(2). The Court’s reasoning was practical: a company receiving millions of complaints cannot be made the judge of which ones are legitimate.
That reading set up a simple bargain. Platforms would not be treated as publishers. In return, they would act promptly on lawful orders. Deciding what is unlawful stayed with courts and the State, through a process that could itself be tested in court.
How the terms have been rewritten:
The IT Rules, 2021 kept the immunity but attached a long list of conditions to it. Large platforms had to appoint resident compliance and grievance officers, publish monthly compliance reports and, for messaging services, be able to identify the “first originator” of certain messages. Challenges to the Rules were transferred to the Delhi High Court in March 2024.
Since then, the pace has picked up:
- The fact-check unit: A 2023 amendment let a government-notified unit flag “fake, false or misleading” information about government business. The Bombay High Court struck it down in 2024 in Kunal Kamra v. Union of India. In March 2026 the Supreme Court admitted the Centre’s appeal, said it raised questions of paramount importance, and declined to stay the High Court’s ruling.
- The Sahyog portal: In September 2025 the Karnataka High Court dismissed X Corp’s challenge to Sahyog, the portal through which agencies across the country send takedown notices. The single judge held that X, as a foreign company, could not claim the Article 19 right, which belongs to citizens. X appealed, and in July 2026 the Supreme Court stayed the appeal proceedings in the High Court.
- Who can order a takedown: An amendment notified on 22 October 2025, in force from 15 November, says takedown intimations must come from an officer of at least Joint Secretary rank (or DIG, for police), must give reasons and the legal basis, and are reviewed every month by a Secretary-level officer.
- Deepfakes and the three-hour clock: Rules notified on 10 February 2026, in force from 20 February, define “synthetically generated information” and require it to be labelled. They cut the time to comply with a lawful takedown order from 36 hours to three, and to two hours for non-consensual intimate imagery, including deepfake nudity. Grievances must now be resolved in seven days instead of fifteen.
- The draft Second Amendment: On 30 March 2026, MeitY proposed making compliance with its advisories, SOPs and directions a condition of safe harbour. It also proposed treating users who post news and current affairs content beyond a threshold as publishers. This came after the Supreme Court, in the India’s Got Latent matter, pressed the Centre in November 2025 to frame rules for user-generated content; the Court eventually closed that case in August 2026 without framing any guidelines of its own. The draft has drawn strong objections and has not been notified.
One pending case shows where this is heading. In the Delhi High Court, in a petition over journalist Rana Ayyub’s posts, the Centre has argued that X is “liable to lose” safe harbour because it did not act after receiving police notices. X’s reply is that blocking must follow the Section 69A procedure and that a private platform is not “State” under Article 12.
Taken together, these changes move India away from the Shreya Singhal bargain. Immunity is no longer the default for a platform that obeys court orders. It is becoming a licence that must be renewed through constant compliance with executive instructions.
Reading it through the Constitution:
Three provisions do most of the work, and each one asks a different question of these rules.
Article 19(1)(a), read with 19(2), allows speech to be restricted only by law and only on the listed grounds. “Misleading information about government business” is not one of them, which goes a long way to explaining why the fact-check unit failed in Bombay. Every new category of prohibited content has to be traced back to that list.
Article 21, after K.S. Puttaswamy (2017), protects privacy, and any intrusion must meet the tests of legality, necessity and proportionality. Traceability, mandatory metadata on AI content and longer data retention all have to clear that bar.
Article 14 strikes down unguided executive power. A three-hour deadline with no hearing before the decision, a review carried out within the same authority that issued the order, and advisories that bind without being notified as law are all open to that challenge.
There is also a gap the law has not closed: the user. The person whose post is removed is rarely told why, rarely heard, and has no obvious forum to contest it. Rule 3(1)(d) notices and Sahyog requests go to the platform, not the author. Platforms tend to comply, because the price of refusing is their immunity. X told the Karnataka High Court that it received 29,118 government removal requests between January and June 2025 and complied with 26,641 of them, about 91 per cent.
Then comes the question of rights against private companies. In Kaushal Kishor v. State of Uttar Pradesh (2023), a Constitution Bench held that rights under Articles 19 and 21 can, in principle, be enforced against non-State actors. What that means for a platform that suspends an account or downranks a post is still unsettled, and it is exactly the kind of question Indian courts will have to answer in the next few years.
The data protection front:
The Digital Personal Data Protection Act, 2023 finally has its Rules, notified on 13 November 2025 and coming into force in phases up to May 2027. For social media companies, the heaviest obligation concerns children.
Under the Act, a child is anyone under 18. Platforms must obtain verifiable parental consent before processing a child’s data, checked against identity or age details they already hold, details the parent volunteers, or tokens from authorised providers. Tracking, behavioural monitoring and targeted advertising aimed at children are prohibited outright.
States want to go further. Karnataka announced a ban on social media for under-16s in its budget on 6 March 2026, Goa and Andhra Pradesh have discussed similar steps, and the Centre is reported to be considering its own age limit. These proposals raise a real question of legislative competence, since intermediaries are regulated under a central statute. They also carry a privacy cost that is easy to miss: to keep children out, a platform has to verify the age of every user, which means collecting more identity data from adults as well.
How other democracies are handling it:
No major democracy has a settled answer, but their approaches differ in ways India can learn from.
The European Union: The Digital Services Act keeps conditional immunity but adds duties of process: notice-and-action systems, a statement of reasons to the affected user, internal appeals, out-of-court dispute settlement, and independent audits for the largest platforms. Its first fine, €120 million against X in December 2025, was for transparency failures (the paid blue tick, the advertising repository and researcher access to data), not for any particular post.
The United Kingdom: The Online Safety Act 2023 puts risk-assessment and child-safety duties on services, with Ofcom as the enforcer. Its early penalties have mostly targeted age-verification failures, including a £1.35 million fine in February 2026, and it has since fined a suicide forum £950,000 under the illegal-content duties.
The United States: Section 230 of the Communications Decency Act still shields platforms from most liability for user content. It is under pressure: bipartisan bills to repeal it or set an expiry date were introduced in both Houses of Congress in 2025 and 2026.
Australia: The bluntest approach of all. Since December 2025, platforms must keep under-16s off their services or face penalties of up to A$49.5 million.
The useful lesson for India lies in the design of the DSA more than in its fines. It regulates how decisions are made, explained and appealed, and it gives the user a place in that process.
What a workable Indian framework could look like:
The government’s position deserves a fair hearing. Platforms earn money from engagement, and they have often been slow to act on content that causes real harm. The Karnataka High Court made a pointed observation in the X case: the company complies with a regulatory regime in the United States but resisted the equivalent in India. Victims of deepfake abuse, child sexual abuse material and coordinated harassment cannot wait weeks for a remedy.
The question is how to get that speed without giving up the safeguards the Constitution requires. In my view, five changes would help:
- Put the core duties in a statute: Whether a platform keeps its immunity should depend on obligations set by Parliament, not on advisories and SOPs. The long-discussed Digital India Act is the natural place for this.
- Tell the user: A takedown made at the government’s request should come with notice to the person who posted, the reasons, and a way to challenge it, with narrow exceptions for national security.
- Separate the person who orders from the person who reviews: A monthly review by a Secretary in the same department is not independent oversight. A body outside the issuing authority, or time-bound judicial review, would be.
- Publish the numbers: Platforms already file monthly compliance reports. The State should publish matching data on orders issued, by agency and by legal ground.
- Match the clock to the harm: Two or three hours is right for intimate imagery and child abuse material. Applied to political speech or satire, the same deadline pushes platforms to remove first and think later.
None of this weakens enforcement. If anything, orders that follow a fair process are far more likely to survive judicial review, which is where many of these rules will eventually be tested.
The question that remains:
Platforms are not neutral pipes, and they should answer for how they design and run their systems. But the right at stake belongs to the citizen. A framework that makes platforms more obedient, without making the process any fairer, risks moving decisions about speech to a place where no court, journalist or user can see them.
Two cases will shape what comes next. The Supreme Court’s hearing on the fact-check unit and the Sahyog dispute will together decide how much of the Shreya Singhal bargain survives.
We would like to hear from people on every side of this: in-house counsel, policy teams, journalists, creators and those in government. Is a three-hour takedown window workable in practice? And should the person whose post is removed have a right to be heard first?
Disclaimer: This article is for informational purposes only and does not constitute legal advice. The content may not reflect the most current legal developments and is not guaranteed to be accurate, complete, or up-to-date. Readers should consult a qualified legal professional before taking any action based on the information provided. The authors and publishers disclaim any liability for any loss or damage incurred as a result of reliance on this article. This article does not create an attorney-client relationship.
