On October 1, 2026, Justice Madhav J. Jamdar of the Bombay High Court granted actor Samantha Ruth Prabhu interim protection against AI deepfakes, morphed images, pornographic content, impersonating chatbots and unauthorised merchandise. Per MediaNama's report, the court found a "very strong prima facie case" and described the misuse, discovered in the third week of July 2026, as widespread, organised and systematic. The part that matters for policy is not the takedown of identified items. It is that Google, Amazon and other platforms must also remove "further material of a similar nature" when Prabhu or her representative notifies them in writing.
The strongest case for the order
The case for this order is strong, and it should be stated fairly. The harms are severe: AI-generated pornographic videos, chatbots holding sexually suggestive conversations in her name, voice clips sold on soundboard sites and counterfeit merchandise. Deepfakes are cheap to reproduce. A victim who must return to court for every re-upload is chasing a problem that regenerates faster than any docket can move. LiveLaw's coverage notes that one chatbot operator voluntarily disabled its AI character after receiving notice of the suit, which shows that early notice can work without litigation. Courts exist to fix gaps like this one.
The order also contains a safeguard. Platforms may communicate objections with reasons, and unresolved disputes can return to the court. That is a meaningful check, and it distinguishes this order from a blank cheque.
Where it collides with settled law
India's intermediary framework rests on a 2015 Supreme Court decision. In Shreya Singhal v. Union of India (24 March 2015), the Court read down Section 79(3)(b) of the IT Act so that "actual knowledge" means a court order or a notification from the appropriate government, not a private complaint. LiveLaw's summary of the judgment records the reasoning: intermediaries receiving millions of requests cannot be asked to decide which are legitimate.
A court order that makes a plaintiff's written notice binding for future content moves part of that adjudication back onto platforms, the very burden the Court wanted to avoid. MediaNama flags this tension. The order does not necessarily conflict with Shreya Singhal, since a court is the source of the obligation. The practical effect, though, is that an interested party decides what counts as "similar" and the platform decides whether to dispute it.
The tension is sharper because the same court spoke differently days earlier. Another Bombay High Court bench, hearing a defamation suit by the Serum Institute of India, held that platforms can be directed to remove only specific content the court has identified and cannot be made to act as censors of future posts. Two benches of one court have taken different positions on forward-looking takedown duties, and platforms operating in India now face real uncertainty about which applies.
The statutory backdrop makes the stakes higher
This is not a legal vacuum. The Information Technology (Intermediary Guidelines and Digital Media Ethics Code) Amendment Rules, 2026, notified by MeitY on February 10, 2026 and in force from February 20, already regulate "synthetically generated information." According to Hogan Lovells' analysis, permitted synthetic content must be prominently labelled and carry permanent provenance metadata. Intermediaries must act on government or court orders within three hours, down from 36. Platforms that knowingly permit or fail to act against violating synthetic content risk losing their compliance protections.
Combine that with a notice-triggered duty and the incentives are clear. A platform facing a three-hour clock and the risk of losing safe harbour has little reason to examine a borderline notice carefully. The cheapest response is to remove first and ask questions never. Over-removal falls hardest on legitimate speech: parody, satire, commentary and criticism involving a public figure, none of which a personality-rights claim should automatically defeat.
A proportionate path
None of this argues against protecting Prabhu. It argues for building the procedure around the order so it scales beyond a famous plaintiff with a capable legal team. Four elements would help:
- Specificity in notices. Each notice should identify exact URLs or account handles and explain why the item is "similar" to adjudicated content, not just assert it.
- A real objection window. The reasoned-objection right is only useful if platforms can use it without losing safe harbour while a dispute is pending.
- A carve-out for expressive use. Parody, news reporting and criticism should be expressly outside the notice mechanism.
- Review of repeat orders. Courts that issue forward-looking relief should build in short review dates so that the scope of "similar" is tested against real takedown data.
The alternative is a patchwork in which each celebrity obtains a bespoke order, each order defines "similar" differently, and platforms rationally over-comply. That is bad for creators, bad for AI developers building lawful voice and likeness tools, and ultimately bad for the victims, because blunt systems are easier to game with false notices.
What to watch
The order is interim, and the Serum Institute decision means a larger bench or the Supreme Court may eventually need to reconcile the two lines. Until then, the Prabhu order shows what a workable model looks like if it is paired with specificity, objection rights and speech carve-outs. Without them, it risks turning a legitimate remedy for deepfake abuse into a general-purpose removal tool.