Whenever a protest spirals into unrest and youth migrate from feed to street, you—Meta—retreat behind a familiar line: that you are merely a platform, not a publisher, making no editorial choices at all. This defence has eroded; for Indian law never granted the unconditional sanctuary found elsewhere in the world.
Dear Meta,
You manage Instagram in India on the implicit consent of some 400 million users, most of them young, and you decide, through opaque weights and undisclosed systems, what they perceive; who they connect with; the very shape of the world outside their window. These impulses originate in Menlo Park—rippling outward into Mumbai-Delhi-Manipur-Kashmir. Whenever a protest spirals into unrest and youth migrate from feed to street, you retreat behind a familiar line: that you are merely a platform, not a publisher, making no editorial choices at all.
This defence has eroded; for Indian law never granted the unconditional sanctuary found elsewhere in the world. Section 79 of the Information Technology Act offers safe harbour only to intermediaries exercising due diligence—not as an automatic courtesy. The IT Rules of 2021 impose transparency obligations, a grievance mechanism, and a specific duty towards children. The Constitution guarantees every citizen—including teenagers scrolling Reels at midnight—the right to speak freely; assemble peaceably; live with dignity. None of this is abstract, it is tested daily by the architecture you have built.
So here are 15 questions. They are neither hostile nor rhetorical. Each is specific enough to be answered with a document or a number, rather than some polished press release.
• One. Your feed ranking optimizes a weighted mix of predicted actions—dwell time, shares, saves, replies, follows. Since safe harbour under Section 79(2)(b) depends on due diligence, have you ever formally measured what share of political or protest content reaches Indian users through algorithmic recommendation rather than chosen accounts? If so, disclose the ratio for civic-news-political content in states seeing youth-led protests these past two years. If not, explain how this opacity squares with a statutory due-diligence standard.
• Two. Before ranking begins, approximate nearest-neighbour retrieval pulls candidate posts from a vast pool. For a teenager who never engaged with protest content, what is the chance that mobilization enters her pool through similarity to fashion-music-cricket? And does your compliance reporting under Rule 4(1)(d) capture this pathway at all, or is the least understood gate in the system simply left out of the transparency you are required to provide?
• Three. Your model predicts probabilities of shares, replies, even reports; high-arousal content is known across the industry to lift these scores. Have you studied whether protest imagery showing confrontation or violence outperforms peaceful imagery among Indian users specifically? If so, what correction exists in the value model to check that bias? If never looked, do you accept this gap is a systemic risk under the accountability regime the proposed Digital India Act would impose?
• Four. Reels and Stories carry documented cascade dynamics; content crossing a velocity threshold is thrown into exponential distribution. Have you modelled a reproduction number for protest content in India? What is the threshold where the algorithm switches from passively carrying to actively amplifying? Who authorized that number? If a Section 69A blocking order arrives after cascading, do you accept that the amplification has already done irreversible harm?
• Five. Quote-posting spins up new documents; it slips past deduplication filters, letting inciting captions ride on peaceful viral posts. Under Rule 3(1)(b), which bars content harmful to children, what stops quote-posts from becoming the main vector for escalating rhetoric among 13-to-18-year-olds who reshare most? And how do you distinguish legitimate counter-speech from harmful hijacking in Hindi-Tamil-Bengali-Punjabi-Telugu, not only in English?
• Six. When most protest content in a young user’s feed comes from accounts she never chose to follow, do you still maintain the algorithm merely reflects interests rather than constructing political awareness? Because if it constructs that awareness, it exercises something close to editorial power over speech protected under Article 19(1)(a), subject to reasonable restrictions of Article 19(2)—and the comfortable distance between intermediary and editor collapses. Do you accept you then bear editorial responsibility, not mere intermediary status?
• Seven. Your guidelines prohibit violence and incitement; yet the line between documenting state violence and glorifying it is context-dependent. As a Significant Social Media Intermediary, you must keep a Grievance Officer resident in India under Rule 4(4). How many reviewers per thousand cases speak the relevant Indian language and are trained to distinguish self-defence footage from armed mobilisation in Kashmir-Manipur-Punjab? What is the false-removal rate for protest documentation? Does it stay internal or reach the Grievance Officer?
• Eight. You deny shadow-banning, even as documents refer to reduced distribution for borderline content. Rule 4(1)(a) requires monthly compliance disclosure; what is the median cut for protest posts that break no rule but get flagged by integrity classifiers? Is this reduction applied evenly across the political spectrum? What audit exists—published, not merely conducted—for ideological bias in the classifiers making that call?
• Nine. You draw identity and behaviour signals across WhatsApp, Facebook and third-party ad-tech partners. For protest mobilisation in India, do you fuse those signals—a WhatsApp group forming alongside an Instagram reshare pattern—to predict offline violence? If so, what stops that fusion from being turned towards pre-emptive suppression of a lawful assembly protected under Article 19(1)(b)? If not, what technical justification exists for leaving data unused, given Rule 3(1)(e) imposes a duty not to host incitement?
• Ten. The proposed Digital India Act contemplates mandatory algorithmic audits for high-risk systems. Has any Indian researcher, regulator or public body ever seen Instagram’s ranking weights or A/B branches for protest content in India? If not, what justifies withholding the blueprint of a system shaping public life, and do you accept that under the coming Act this opacity may itself constitute non-compliance?
• Eleven. When your algorithm demotes a user’s protest post or restricts an account on a classifier’s prediction, do you treat that as an automated decision under the Digital Personal Data Protection Act, whose Section 12 grants a right to information about processing? Are Indian users ever told their content’s reach was determined by prediction rather than any rule they broke? If not, what legal reasoning distinguishes your feed ranking from automated processing under Indian data-protection law?
• Twelve. Sections 153A and 505 of the Penal Code criminalise speech stirring communal or public disorder. Have you conducted a systemic risk assessment of Instagram’s role in amplifying content crossing that line during youth-led protest? What mitigations are mandated internally, who is accountable under Indian criminal law? If a court finds your algorithm amplified content violating Section 153A, do you accept that liability may attach to the platform, not only the user?
• Thirteen. Your own research found Instagram worsened body image for nearly a third of teen girls. Has an equivalent study of the regret gap—engagement followed by reported distress—been run for Indian youth exposed to protest violence on the platform? If so, what architectural change followed, beyond warning labels, given that Rule 3(1)(b) makes child protection a positive duty rather than a suggestion?
• Fourteen. When protest-related violence is linked to content that spread on Instagram, your standard answer is that users broke the rules. Yet Indian courts, following Shreya Singhal, left the door open for platforms actively participating in unlawful acts to lose their immunity. Do you accept that the design of the recommendation system—its blindness to emotional valence, its cascade thresholds, its retrieval boundaries—is itself an active choice with consequences? If not offer the alternative account clearing the architecture of responsibility under Indian tort and criminal principles.
• Fifteen. The value model governing what several hundred million Indians see is tuned by product teams thousands of kilometres away, with almost no Indian democratic oversight. Under Article 14, bias that disproportionately suppresses or amplifies political content may amount to arbitrary action; under Article 21, protecting dignity and privacy, the stakes reach further still. Will you support an independent, statutory Indian body empowered to review and, where necessary, order changes to ranking weights for content categories of clear civic consequence? If not, what mechanism do you propose for democratic accountability over a system shaping collective action across 400 million lives—or do you simply propose none?
These fifteen questions are not an indictment; they are an invitation, and the distinction matters. India’s legal architecture already says what it needs to say—safe harbour is conditional, transparency is required, child safety is non-negotiable, fundamental rights do not bend to convenience—and your answer will be measured against that architecture, not against your own comfort. What hangs in the balance is not one company’s reputation but whether four hundred million Indians, many still in school, will go on being sorted and shown the world by a system nobody outside Menlo Park is permitted to see inside.
The law is only beginning to catch up. The Constitution was here before either of us. The Citizens of India
*Brijesh Singh is a senior IPS officer and an author (@brijeshbsingh on X). His latest book on ancient India, “The Cloud Chariot” (Penguin) is out on stands. Views are personal.
