By BarathVector Editorial — 2026-08-14

A convincing fake needs minutes to travel. A correction needs permission, attention and luck.

That imbalance is why a visible label on synthetic media is necessary—and why it is nowhere near sufficient. By the time a platform marks a viral clip as manipulated, the alleged confession has reached family groups, television panels and political speeches. The target may win the forensic argument and still lose the election, the job or the trust of a community.

India's response to deepfakes should begin before upload, move faster after harm and preserve evidence from the first moment. The objective is not merely to tell viewers that a file may be artificial. It is to make authentic material easier to verify and malicious fabrication costly to distribute.

Labels solve only the visible part

India's 2026 amendments to the Information Technology rules require platforms to take measures against unlawful synthetic content, preserve traceable metadata and apply prominent labels in relevant circumstances. A government explanation of the changes also says intermediaries must act within three hours when an authorised government or court order requires removal.

Those are useful obligations. Viewers deserve to know when a realistic voice, image or video was generated or materially altered. Traceable disclosures make casual deception harder. Rapid action can limit a second wave of circulation.

But a label can be removed by cropping, screen recording or re-encoding. A malicious uploader will not honestly declare the file synthetic. Automated detectors make mistakes, particularly after compression and editing. And even a perfect label cannot reach copies already saved on private devices.

The deeper problem is not that every fake lacks a warning. It is that authentic media often arrives without a trustworthy history.

Provenance must begin at capture

Modern cameras and publishing systems can attach cryptographically signed information about when media was captured, which device or organisation handled it and what edits followed. The C2PA specification provides an open technical standard for such content credentials.

Provenance is not a truth machine. C2PA's own explainer warns that a valid history does not prove the scene itself was honest. A staged video can be authentically recorded. A genuine file may lack credentials because it came from an old or inexpensive device. Absence of a signature must never be treated as proof of fakery.

Used properly, provenance answers narrower and valuable questions: who claims to have created this file, has it changed since, and is the chain intact? That can help a newsroom verify official footage, a court preserve exhibits and a citizen distinguish an original statement from an edited copy.

India should require capture-to-publication credentials first where institutional responsibility is clearest: election authorities, police body cameras, government press offices, public broadcasters and registered political advertisements. Newsrooms and device makers should be given interoperable tools and a transition period. The standard should remain open so that verification does not depend on one foreign platform.

Detection is evidence, not a verdict

The public is routinely shown detector scores as if they were laboratory certainty. They are not. The US National Institute of Standards and Technology's 2026 deepfake evaluation reported steep performance degradation under operational conditions. Compression, unseen generators and ordinary post-processing can make a benchmark winner unreliable in the wild.

India needs accredited forensic laboratories, documented methods and repeatable handling of disputed files. A platform classifier may justify slowing distribution while a claim is examined. It should not by itself convict a creator or declare a damaging recording authentic.

The Bharatiya Sakshya Adhiniyam recognises electronic records and provides procedures for their admissibility, including certificate and hash requirements. Deepfake policy should strengthen that evidentiary chain: preserve the original file, its cryptographic hash, upload records, platform transformations, detector versions and expert findings. Screenshots of a social-media post are not enough.

The strongest countercase

Provenance systems create risks of their own. Persistent device identities can expose journalists, dissidents and witnesses. Mandatory signing may give governments a convenient way to dismiss genuine anonymous footage. Smaller creators may be excluded if trusted credentials become an expensive badge.

Those objections should shape the system. Public verification can reveal that a credential is valid without exposing a person's civil identity. Anonymous and pseudonymous publication must remain possible. Courts and newsrooms should evaluate unsigned material by ordinary corroboration rather than assuming it is false. Open-source tools and free verification should prevent a two-tier media system.

Privacy cannot be an afterthought attached to an authenticity scheme. Yet privacy concerns do not justify leaving every citizen alone against industrial fabrication.

Remedy must match the speed of harm

India needs a fast civil process for impersonation and intimate or reputational deepfakes. A victim should be able to secure preservation of evidence, temporary limits on recommendation and forwarding, and an urgent hearing without first proving the entire underlying case. Platforms should notify accounts that shared the material when a final determination finds it fabricated.

Election periods require an even shorter clock. The Election Commission has already warned parties against deepfakes and directed prompt removal of unlawful content. That should be backed by a standing technical cell, a published escalation route for candidates and citizens, and archived decisions that explain what evidence was considered.

Penalties should distinguish satire and disclosed creative work from impersonation intended to deceive. The law should punish malicious conduct and provable harm, not the mere use of generative tools. Broad criminal language would chill parody, cinema and political criticism while serious operators move offshore.

Trust is built forward

The most damaging effect of deepfakes may be the liar's dividend: guilty people dismissing genuine evidence as AI. A society that treats every recording as suspect does not become sceptical; it becomes governable by whoever has the loudest distribution network.

Labels help viewers pause. Provenance helps investigators reconstruct. Forensics helps courts decide. Rapid remedies reduce the time in which a lie can operate without resistance. No one measure restores trust on its own.

India should therefore judge its synthetic-media regime by a harder test than the number of labelled posts: can an ordinary person establish what happened, preserve admissible evidence and obtain meaningful relief before the falsehood becomes permanent public memory?

A deepfake label may arrive on the screen. Reputation lives outside it. Policy must reach both.