Globally, there is continuous uploading of photos, videos, and audios through social media on a minute-to-minute basis. Despite the vastness of the human labor force, it is impossible for them to go through each piece that is uploaded. Thus, many social media sites have resorted to automatic systems to analyze such content and act against any misinformation, privacy breaches, and security concerns. India is now one of the countries where social media platforms such as Facebook and Instagram automatically remove content that these systems flag, and takedown notices are handled almost immediately. The Ministry of Electronics and Information Technology (MeitY) has notified the Information Technology (Intermediary Guidelines and Digital Media Ethics Code) Amendment Rules 2026 to curb synthetically generated information (SGI) or deepfakes, which is the genuine need of the hour, but the manner in which the government is compelling this automation is what makes the exercise dangerous. 

The shift from Human to automated censorship – Indian Framework

The Supreme Court in Shreya Singhal v. Union of India struck down Section 66A of the Information Technology Act, 2000 entirely for violating the fundamental right of freedom of speech and expression under Article 19(1)(a) of the Constitution. The court decided that an internet intermediary's “actual knowledge” of content under Section 79(3)(b) of the IT Act needs a court order or a government notice. Also, any takedown that happens must be covered under Article 19(2). In the case of Anuradha Bhasin v. Union of India extended accountability to internet access, holding that speech on the internet is protected under Article 19(1)(a). The court also said that any limit, on that speech must follow the idea of proportionality.

Last year, the government introduced the “Sahyog Portal” to automate the process of sending notices to intermediaries. The Appropriate Government or its agency under the IT Act 2000 uses this portal to help remove or block access to any information, data, or communication link used for an act. The Information Technology (Intermediary Guidelines and Digital Media Ethics Code) Amendment Rules, 2026, came into force on 20 February 2026 and target synthetically generated information (SGI), deepfakes, and other AI-altered audio-visual content.

According to the new Rule 3(3) requires intermediaries to use appropriate technical measures, including automated tools to stop the unlawful use of SGI. Rule 3(3) also mandates that high‑risk SGI categories be prohibited, that non‑prohibited SGI content carry labels and provenance information, and that such labels be tamper‑proof. More significantly, large social‑media intermediaries must ask a user whether the uploaded content is SGI and must verify that answer with automated detection tools before the content is published. Rule 4(4) has changed from an “endeavour” to a duty. The 2026 Rules drastically cut down the response times. Platforms now have just three hours to act on most takedown orders, down from 36 hours. For grievances about content removal, the window shrinks from 72 hours to 36. And for sensitive material like nudity, sexual content, morphed images, or impersonation, platforms must respond within two hours instead of 24. With millions of posts uploaded daily, these tight deadlines leave no room for human review, and only automated systems can move fast enough.

Why Automation Makes the Problem Worse

There are four features of automated moderation that make the constitutional question worse than a single takedown order. First, over-removal: to maintain safe-harbor immunity, platforms use AI-based models to filter content, which is something perfectly safe and lawful and falls under Article 19(2) to avoid false negatives. Second, opacity: users are not ever informed as to which rule was violated or what triggered the removal, and the details of how the algorithm scores content are not shared with the public. Third, lack of contextual understanding: automated processes struggle with satire, regional language, and political speech, all of which, according to Singhal and Bhasin, deserve special protections. Fourth, scale and deadline combined: with removal decisions due in as little as three hours, there is no real chance for human review, and one officer's view of what counts as public mischief can end up affecting content across the entire country thanks to how platforms comply nationwide.

Deepfakes are a problem, and misinformation can have serious effects that deserve regulation. This does not change the government's reason for regulating. A policy that uses fear and cuts every deadline to a tiny window that only machines can meet is not the same as an accountability‑based approach. Genuine internet safety will come from communication about why content was removed, from digital literacy programs, and from judicial accountability of the police and algorithms that enforce these regulations not from deadlines that make oversight impossible by design.

The European Union - a workable model

The European Union's Digital Services Act (DSA) takes a different structure than that of India; it does not attempt to tell platforms what speech is allowed, but it makes sure there is a clear process when content is taken down or limited. Article 17 provides that a service that hosts content must give anyone affected a specific reason whenever they take down or limit something. Article 20 provides that platforms must provide an internal complaint-handling system, free of charge, for at least six months after the decision. Article 21 further adds that when a dispute remains unresolved, an out-of-court settlement should be done by a body independent of the platform. Article 14 requires that a platform's terms of service be applied in a diligent and proportionate manner with due regard to the fundamental rights of users under the EU Charter. For the bigger platforms, Articles 34 and 35 additionally provide for a systemic-risk assessment to cover the impact of their algorithmic systems on freedom of expression and media pluralism, together with mitigation measures.

What matters most is that the government doesn't need to identify lawful speech for algorithmic moderation to be constitutionally permissible. What is required is for the State to ensure that every user is informed, provided a way to challenge the decision, and provided an external forum to the platform to examine the decision. It is this very combination that is currently missing from India's Grievance Appellate Committee framework, as well as the new 2026 Amendment Rules process.

Four practical reforms could improve India's content moderation system.

First, every time a post is removed, the user should receive a clear notice explaining exactly which rule or automatic signal caused the removal. This is not something that should be done but is a simple fairness as decided in the case of Maneka Gandhi v. Union of India.

Second, appeals should go to an independent body, not a Grievance Appellate Committee staffed entirely by the executive, but one with judicial or independent membership empowered to review the content.

The three‑hour removal deadline should not be applied blindly to all content. Political speech and journalism deserve at least a brief window for a human to review before an algorithm makes a final decision. This follows from the proportionality doctrine in K.S. Puttaswamy v. Union of India, (2017), applied to speech restrictions in Anuradha Bhasin. A restriction must use the least restrictive means available, and a blanket, content‑blind deadline that treats a deepfake and a political cartoon the same cannot be seen as narrowly tailored to a purpose.

Fourth, platforms and the government must publish transparency reports that show how often automated systems make mistakes and how many takedowns are reversed on appeal. Without this data, there is no way to verify whether the system is being used proportionately. Courts, researchers, and citizens need facts, not assurances.

Underlying all four reforms is a deeper point: even with safeguards, three-hour deadlines create a chilling effect. When decisions must be made that fast no platform, officer or court has time to test whether a restriction is legal before it takes effect. This is the problem that the Supreme Court identified in Shreya Singhal: vague or immediate restrictions suppress lawful speech

The algorithm has become a censor operating outside constitutional scrutiny of Article 19, too automated to be treated as a human editor, too consequential to be left unregulated, and now too fast to be meaningfully supervised. India should combine Shreya Singhal's demand for accountable, reasoned decisions with the European Union's guarantees of notice, appeal, and independent review.