On 15 September 2026, the Delhi High Court (Justice A.J. Bhambhani) modified an earlier interim injunction in Dr. Aniruddha Dhairyadhar Joshi v. John Does Ashok Kumars & Ors., CS(COMM) 178/2026 (Joshi). The plaintiff is a spiritual leader, and the suit targets AI-generated deepfake videos of him. The earlier order of 24 February 2026, passed by Justice Tushar Rao Gedela, directed Google, Meta and X to remove identified deepfake content, and also content “similar to” it. The platforms objected: if a plaintiff can simply call new content “similar”, can an intermediary be compelled to remove it without any court order about that content? That put Section 79 of the Information Technology Act, 2000 and Shreya Singhal v. Union of India in issue.
The September order sidesteps the safe-harbour question by splitting the remedy: provisional enforcement against identical content, and platform-policy-based removal for similar content. This blog argues that the split is a pragmatic compromise. It works for true duplicates, but it pushes “similar” content into private moderation with little supervision and little protection for the uploader.
Why the case matters
The order matters less for what it decides than for what it designs. The suit pleads copyright, personality and publicity rights and engages intermediary liability, the classic tension between protecting right holders and not turning intermediaries into universal monitors. The order decides none of these rights; it delegates verification to platforms. Google applied under Order XXXIX Rule 4 read with Section 151 CPC to vary the February direction. The plaintiff and Google then filed a joint application for disposal on agreed terms. After also hearing Meta, the court substituted paragraphs 38(ii) and 38(iii) to resolve the concerns of both sides. As an interim modification reached largely by agreement, it is not a final adjudication and carries limited precedential weight.
Re-uploads and dynamic relief
The February order required removal within 48 hours of a complaint, for content uploaded from IP addresses in India. The plaintiff’s case was practical: a removed deepfake reappears on another URL, account or platform, and a fresh application each time makes the injunction ineffective.
Courts issue “John Doe” (or “Ashok Kumar”) orders against unidentified defendants and, in piracy cases, dynamic injunctions. In India the dynamic injunction comes from UTV Software Communication Ltd. v. 1337X.to, 2019 SCC OnLine Del 8002, which let plaintiffs add mirror, redirect and alphanumeric versions of already-blocked rogue websites without a fresh suit. The “dynamic+” variant (Universal City Studios v. Dotmovies.baby, 2023) resembles the identical-content limb below. The danger is that if “similar” becomes too broad, a platform must decide whether millions of items infringe someone’s rights, and a judicial remedy becomes private censorship.
The platforms’ objection
Google and Meta argued that the direction was overbroad because it required them to act on content about which no “actual knowledge” could be attributed through a court order, as Shreya Singhal contemplates. They relied on Section 79, the 2021 Intermediary Rules, Shreya Singhal and the Division Bench decision in MySpace Inc. v. Super Cassettes Industries Ltd., 2016 SCC OnLine Del 6382. They said they could not remove content without a specific court order unless acting under their own content policies. The order records both positions without deciding the safe-harbour question.
The s.79(1) exemption for third-party content is subject to sub-sections (2) and (3), and under s.79(3)(b) it is lost if, on receiving actual knowledge, the intermediary fails to act expeditiously. Shreya Singhal, (2015) 5 SCC 1, best known for striking down s.66A, read down s.79(3)(b) and its corresponding rule (now Rule 3(1)(d) of the 2021 Rules): actual knowledge means a court order or government notification, not a private complaint. MySpace harmonised Sections 79 and 81 with copyright law and stressed that relief against intermediaries must identify infringing content specifically.
Two points follow. First, s.79 is a shield against liability, not a limit on a court’s injunctive power, and Shreya Singhal itself treats a court order as the paradigm trigger. The platforms’ real objection was reach an order extending to content no court has examined. Second, MySpace is a copyright decision, whereas personality and publicity rights are judge-made and have no statutory notice-and-takedown scheme of their own. The intermediary argument may therefore sit differently across the limbs of this single injunction.
The revised framework
Relying on Home Box Office Inc. v. Streamzy.to (HBO, below), the court created separate mechanisms.
Identical content. If the plaintiff finds URLs, posts, accounts or handles carrying content identical to the infringing content covered by the February order, he may inform Google, Meta or X on affidavit, with particulars and supporting documents evidencing identity. The intermediary must technically verify whether the content is identical and, if so, enforce the injunction against it pro tem (provisionally). Simultaneously, the plaintiff must apply to implead those URLs, posts, accounts or handles, and the court will consider the application.
Identical content is easier to verify: if a new upload is the exact same video as the original deepfake, technical comparison can establish that. “Similar” could mean the same subject, allegation, person, visual style, words or message, which requires evaluative judgment. Confining mandatory enforcement to identical content reduces, without eliminating, overbroad takedown risk. But the order sets no deadline for verification or for the court’s decision on impleadment, and says nothing about enforcement if impleadment is delayed or refused. A platform ignoring a verified duplicate risks contempt (Order XXXIX Rule 2A).
Similar content.
For merely similar content, the plaintiff may give information on affidavit, but the intermediary removes it only if it violates its own policies or guidelines framed under the IT Act and Rules, and then within 48 hours of the request. That is not a judicial finding of infringement, and with no verification or impleadment on this limb, the order builds in no court supervision. The practical gap may be narrower than it looks: reports of earlier hearings record the plaintiff noting that Google’s own policies already permit removal of content using a person’s name, image or likeness. But that depends on each platform’s discretion, and the order does not say what follows if a platform fails to remove similar content that violates its policy.
Subscriber information
The February order already required subscriber disclosure; the substituted paragraph 38(iii) restates it and adds a limit. Google, Meta and X must disclose available Basic Subscriber Information (BSI: name, address, phone number, email, IP address and other identifiers) for URLs, posts, accounts and handles connected with the infringing deepfake or impersonated content, uploaded from IP addresses in India, that have been blocked, removed or disabled. BSI for additional or identical URLs under paragraph 38(ii) is disclosed only on specific court directions and also should be password protected.
Identification differs from removal: it affects privacy and may touch people not ultimately found responsible, so disclosure should meet Puttaswamy‘s necessity and proportionality standard. There is also a paradox. Paragraph 38(iii) covers content removed “pursuant to the directions above”, which arguably includes policy-removed similar content with no court check, while BSI for “additional/identical” URLs needs specific court directions. If policy removal counts, the least-verified content may carry the easiest disclosure.
HBO: lineage and limits
In Home Box Office Inc. v. Streamzy.to (CS(COMM) 740/2026, 27 July 2026), the same judge enjoined 30 rogue websites and, applying UTV, directed ISPs and domain-name registrars to technically verify whether a newly found site is a mirror, redirect or alphanumeric version of an injuncted site and, if so, to enforce the injunction provisionally while the plaintiff moves to implead it. Neither the plaintiff nor intermediaries could decide unilaterally what is a rogue website; the court delegated only technical verification, not discretion to block.
Joshi borrows this architecture but differs in two ways. HBO concerns access to websites, handled by infrastructure intermediaries, after a prima facie finding that the sites are rogue; Joshi concerns individual content on hosting platforms, where the underlying rights differ. And whether a domain is a mirror is a different inquiry from whether a video is identical. Because both orders come from the same judge, the second is also no independent confirmation of the first.
Technical identity is not always straightforward
The order provides no test for when modified content stops being “identical”. A file can be compressed, resized, cropped or re-encoded without changing its content, while a short edit can change a video’s meaning or remove what made it unlawful. Three inquiries should stay distinct: technical identity; perceptual or substantive sameness despite format changes; and whether the new material falls within the injunction. Hash matching may miss altered copies, while perceptual tools may produce false positives.
A later upload might reuse the video, put the likeness in a new scene or clone the voice for new statements; these are not legally interchangeable. The 2026 IT Rules amendment (G.S.R. 120(E), notified 10 February, in force 20 February) requires labelling and metadata for synthetically generated information and technical measures against unlawful synthetic content. Such markers may help verification but are no substitute for a legal test.
The rights must also be distinguished
Copyright may protect a photograph or video, but the person depicted does not necessarily own it. A synthetic endorsement may raise passing-off or identity concerns, a fabricated statement defamation, and parody and commentary engage expressive interests. Courts have granted deepfake-specific relief before (Anil Kapoor v. Simply Life India, Delhi High Court, 2023), but new content resembling restrained material does not necessarily infringe the same right, and the order does not say which right each limb enforces.
Grey areas
Platform governance and the Rules. The 2026 amendments set a three-hour clock for removal on court or government orders (Rule 3(1)(d)) and a two-hour window for individual complaints about impersonation, including morphed images (Rule 3(2)(b)). The 48-hour policy window cannot displace either, which raises two questions: what the “similar” limb adds where Rule 3(2)(b) already applies, and whether the three-hour clock governs the identical limb, where the order sets no deadline. Enforcement may also vary by platform policy, and the order provides no transparency or review.
Uploaders. The person whose content is removed is not heard, and the order has no counter-notice or review route. Significant social media intermediaries must notify users of voluntary removals under Rule 4(8), but whether policy-route removals trigger that duty is unaddressed. Article 19(1)(a) interests are barely touched, though parody and commentary can look “similar”.
Drafting gaps. The substituted paragraph 38(ii), as reproduced, omits the India-IP-address qualifier of the original direction, while paragraph 38(iii) retains it; whether that is deliberate is unclear. The order also records Meta as heard but records no appearance for X, and only Google joined the agreed-terms application, though the substituted paragraphs bind all three.
What a tighter model looks like. In Glawischnig-Piesczek v. Facebook (C-18/18, 2019), the Court of Justice of the EU allowed injunctions reaching identical and “equivalent” content if the order specifies what is to be removed, so the host need not make an independent legal assessment. This order identifies the comparator for identical content but not how identity is tested or what makes content “similar”.
Conclusion
The Joshi order is a modest improvement on the February direction. For true duplicates it is a sensible, court-anchored mechanism that, like HBO, leaves the final say with the court. For similar content it hands enforcement to private policy, making platforms judges of similarity and policy compliance. Several fixes would help: define “identical” in technical and perceptual terms; set deadlines for verification and impleadment decisions and say what follows if impleadment fails; give uploaders a review route; require transparency on policy-based removals; and clarify the BSI proviso.
Speedy enforcement matters, but technical similarity alone cannot establish illegality. The legitimacy of such remedies depends on identifying the right infringed, the wrong alleged, and the limits of permissible restriction.
Authored by Mr. Kartik D Verma, National law School of India University