Bombay HC Orders Meta, Google to Pull Gadkari Deepfakes - And Exposes India's Reactive Takedown Regime

The Real Story Isn't the Deepfake

• It's Who Had to Go to Court to Stop It Here's what's actually happening underneath the headline: a deepfake is just AI trained to swap a face, clone a voice or fabricate a scene convincingly enough that a viewer can't tell it's fake without technical help. That part isn't new - face-swap tools have existed for years. What's new is that the barrier to making a convincing one has collapsed. Producing 26 separate fabricated clips linking a Cabinet minister to a policy he doesn't administer and to alleged financial gains he never made, no longer takes a studio. It takes a laptop.

Why the Platforms Didn't Act First This is the

part worth sitting with. Meta and Google didn't proactively detect and remove this content - they agreed to take it down only after being named as defendants in a court order. Under the IT Rules, 2021, intermediaries are required to exercise "due diligence" and act on unlawful content once they have "actual knowledge"

• usually meaning a court order or a government notice. That threshold sounds reasonable until you notice what it actually means in practice: platforms with enormous automated moderation systems for spam, copyright and CSAM largely don't run the same detection-first logic for defamatory deepfakes. Justice Doctor asked the platforms' lawyers this almost directly in court - with all this technology at their disposal, why isn't there a standing policy to flag "patently vile, abusive" content the moment it's uploaded?

Where India Actually Stands

On paper, India has one of the more detailed intermediary liability frameworks globally - the 2021 Rules are more prescriptive than the US's Section 230 model. But detail isn't the same as design. The rules are built around notice-and-takedown, a model inherited from an earlier internet era of pirated files and defamatory blog posts, applied unchanged to a technology - generative AI - that can manufacture novel, hyper-realistic fabrications at a scale the notice-and-takedown model was never built to absorb.

The court's frustration in this case is really a frustration with a mismatch: the harm moves at AI speed, the remedy moves at litigation speed. The exam-relevant point isn't "deepfakes are dangerous"

• everyone already knows that. It's that India's current framework treats every deepfake case as a fresh legal event requiring a fresh court order, rather than building a standing, detection-first obligation into the platforms themselves. A minister with resources got relief in weeks. An ordinary citizen without a lawyer on retainer may not get relief at all - and that gap between who can access the remedy and who the remedy was meant to protect is where this story actually lives.