I4C Orders GitHub Takedown of Bluetooth-based app- BitChat 

On 23 July 2026, the Indian Cyber Crime Coordination Centre (I4C), under the Ministry of Home Affairs, issued Notice to GitHub directing the platform to disable access to three repositories, including the Android application and its release files, within three hours, failing which GitHub risked losing safe harbour protection under Section 79  of the Information Technology Act, 2000 and exposure to criminal liability. The order was issued under Section 79(3)(b) of IT Act, read with Rule 3(1)(d) of the IT Rules, 2021. The Government of India published no copy of the notice and the public learnt of it only through a disclosure by BitChat’s developer, Jack Dorsey, on X. This order is in direct violation of Anuradha Bhasin v Union of India wherein the Supreme Court held that all orders imposing restrictions on fundamental rights, including freedom of expression and the freedom to carry on any trade or profession via the internet, must be published publicly. This ensures that affected individuals and developers can legally scrutinize and challenge them. Furthermore, a three-hour deadline issued near midnight forecloses any meaningful legal assessment or recourse, as per the statement issued by  Internet Freedom Foundation (IFF).

BitChat is an open-source Bluetooth mesh messaging application that allows devices to relay encrypted messages hop-by-hop without reliance on mobile networks, internet connectivity, or centralised servers; it separately supports internet-based communication through the decentralised Nostr protocol. The application requires no account, phone number, or persistent user identifier, and offers end-to-end encryption for private messages. Its Android repository carries a developer-issued warning that the software has not undergone external security review and should not yet be relied upon for sensitive communications.

The order followed an internet shutdown at the Jantar Mantar protest site and parts of central Delhi. I4C’s stated justification was that such platforms could be used to evade lawful surveillance and internet shutdowns, and that the absence of a centralised operator prevented law enforcement from accessing subscriber information or communication logs which poses a threat to sovereignty and public order. The order failed to identify any illegal text, data, or content hosted within the repositories, targeting the app’s underlying decentralized Bluetooth mesh technology instead.  Moreover, it places reliance on Section 79(3)(b), which is ordinarily an intermediary-liability provision. However, here, it is used to achieve what is functionally a content-blocking outcome and thus, it effectively bypasses the more structured procedural safeguards under Section 69A and Blocking Rules 2009.

The order’s practical efficacy has been contested from the outset. GitHub had not taken down the repositories as of the time of reporting, and the application is available on both the Google Play Store and Apple App Store. Commentators have noted that deleting a repository removes only one copy of open-source, forkable code; it neither deletes the application from devices that already carry it nor disables a mesh network that requires no server to function.

BitChat is not the first such tool to draw state attention: China had already ordered BitChat’s removal from Apple’s App Store in April 2026 and had banned Dorsey’s earlier Nostr-based application, Damus, in 2023, making India the second major state to move against the app. Mesh-networking tools of this kind, such as Bridgefy and FireChat, have a documented history of adoption by protest movements during periods of state-imposed connectivity restrictions, precisely because their design objective is continuity of communication when centralised infrastructure is unavailable or deliberately withdrawn.

The BitChat order demonstrates a wider pattern of restricting freedom of speech this year. Reporting and rights-organisation tracking through 2026 have documented account-blocking actions against protest movements and press accounts under Section 69A, frequently without published orders or an opportunity to contest them before the Inter-Departmental Committee; internet suspensions localised to specific protest sites; and a continued reliance on colonial-era and anti-terror statutes against journalists covering contentious subjects. India’s position in international press-freedom indices has declined further this year, with commentary specifically noting a shift in which legal instruments function less as safeguards for journalists and more as tools deployed against them. 

Delhi HC dismisses ANI Media’s Application for an Interim Injunction alleging Copyright Infringement

On 24th July 2026, a single-judge Bench of the Delhi HC delivered the final judgment in an interim application filed by ANI Media Pvt. Ltd. seeking an injunction to restrain OpenAI from using, scraping, storing, or reproducing its copyrighted news content to train OpenAI’s Large Language Models (LLMs) or to generate outputs via its AI application, ChatGPT. ANI founded its suit on two primary grounds: (i) the “training claim” (alleging that the scraping and storage of ANI’s news articles on servers for model training constitutes copyright infringement under Section 51 read with Section 14(a)(i) of the Copyright Act, 1957); and (ii) the “reproduction/output claim” (alleging that ChatGPT memorizes and regurgitates ANI’s copyrighted text in responses provided to users).

The HC heard the parties along with multiple intervenors (news publishers, music industry bodies, think tanks, and AI start-ups) and two amici curiae, one of which was NLSIU’s very own Professor Arul George Scaria (you may find our interview of Professor Scaria on his involvement in the case here). 

In its final ruling, the Delhi HC held that ANI failed to establish a prima facie case of copyright infringement for either claim and that the balance of convenience and public interest weighed heavily against granting an interim injunction to ANI Media. Here is a quick summary of key points covered in this very well-structured and thoughtfully written judgment by J. Amit Bansal:

(A) Location of AI Training Servers Does Not Defeat Indian Copyright Jurisdiction under Section 62

OpenAI raised a preliminary jurisdictional objection asserting that Indian courts lack jurisdiction over the training claim because the training of the LLM occurs on servers located in the United States, arguing that the Indian Copyright Act, 1957 cannot have extra-territorial reach.

The HC firmly rejected this argument on a prima facie view. It used principles from the CPC to establish territorial jurisdiction under the Copyright Act, as ANI’s principal place of business and registered office are in Delhi, and because OpenAI specifically targets/offers its services to users/subscribers across India. 

On the location of servers, the HC held that the storage of data on servers in the US is merely a terminal step in an unbroken chain of events that begins with accessing and transmitting copyrighted content from India. Therefore, the HC refused to sever the training process from the output generation, noting that placing data on overseas servers cannot be allowed to be used as an escape route to evade Indian copyright laws in an era dominated by cloud computing.

(B) ChatGPT’s Outputs Did Not Amount to Substantial Reproduction of ANI’s Copyrighted Expression

The HC examined whether ChatGPT’s generated responses constitute an unauthorized reproduction or substantial copying of ANI’s copyrighted literary works under the Copyright Act. ANI argued that LLMs inherently store and memorize raw training text, leading to the “regurgitation” of verbatim copies when prompted by users. 

The court evaluated technical evidence and submissions from the amici curiae observing that LLMs operate as computational, probabilistic prediction systems rather than permanent repositories of raw text. OpenAI here made an admission that verbatim memorization of training data is an unintended, rare occurrence (often triggered only via specialized “extraction attacks” or “adversarial prompting”). However, the ANI Media failed to establish a case of regurgitation of their content – the nine illustrative articles cited in the plaint to prove output copying were published between August and September 2024. In contrast, the pre-training cutoff dates for OpenAI’s relevant models were April 2022 (GPT-4) and April 2024 (GPT-4o). Because the articles were published after model training had already concluded, ChatGPT could not have memorized them during training.

ANI Media had provided specific output examples, such as an interview with Olympic athlete Neeraj Chopra’s mother, to argue that their copyrighted content was being regurgitated. However, the HC sided with OpenAI who highlighted that standard user prompts produced unique summaries with ChatGPT’s own contextual commentary. Near-identical text was elicited only after ANI deployed deliberate, adversarial prompts instructing ChatGPT to reply “exactly” with what was said. Further, the following points were made with respect to the copyrightability of the impugned content:

1. No Copyright in Facts or News: Citing Feist Publications, EBC v. D.B. Modak, and Akuate Internet Services v. Star India, the Court reiterated that facts, news events, and historical occurrences are not copyrightable per se because protection is strictly limited to the manner of expression and not the idea itself.
2. Ownership of Quotes: Under Section 17(cc) of the Copyright Act, the first owner of copyright in a public speech or oral interview is the speaker, not the reporting news agency (absent an explicit assignment). Thus, verbatim quotes from interviewees cannot constitute infringement of ANI’s copyright.
3. Comparison as a Whole: Relying on R.G. Anand v. Deluxe Films and Star India v. Leo Burnett, the Court emphasized that rival works must be compared as a single entity in their entirety, rather than dissecting isolated sentences or extracts. When comparing ANI’s full articles against ChatGPT’s complete outputs, the Court found no substantial material similarity in expression.

(C) Dynamic Interpretation of ‘Private Use’ and ‘Research’ Extends Fair Dealing to AI Training

To assess whether OpenAI’s temporary storage and processing of data for training LLMs qualifies as fair dealing under Section 52(1)(a)(i) (“private or personal use, including research”), the HC made the following observations:

• Rejected ANI’s argument that commercial entities cannot claim fair dealing under Section 52(1)(a)(i). The Court contrasted Section 52(1)(a) with other sub-clauses [such as Section 52(1)(ad) or 52(1)(n)] where the legislature explicitly inserted the restriction “non-commercial.” Where no such express bar exists, commercial use does not automatically invalidate a fair dealing defense.
• Clarified that the phrase “which is not itself an infringing copy” in the Explanation to Section 52(1)(a) applies specifically to the incidental storage of computer programmes, not general literary works. Regardless, OpenAI accessed ANI’s content lawfully from publicly accessible web pages without breaching paywalls or technical protection measures.
• Relied on the doctrine of updating construction (i.e. statutory language must be interpreted dynamically to account for modern technological advancements) to rule:

1. “private use” is not restricted to individual natural persons but encompasses internal corporate operations where data is processed in a non-public environment.
2. “research” can no longer be limited to human study alone; it extends to machine learning, computational data analysis, pattern extraction, and algorithmic model training undertaken to advance AI systems.

Further, since the use under Section 52 must be “fair,” the HC formulated a three-factor test aligned with Article 9 of the Berne Convention to determine the ‘fairness’ of this emerging technology. This is a greatly innovative approach adopted by the Delhi HC not often witnessed in case law.

1. Limited Use for Training: OpenAI uses stored literary works internally solely to extract non-expressive elements (grammar, syntax, linguistic patterns, and semantic relationships) to train parameters. The raw text is not displayed, published, or distributed to the public.
2. Lack of Economic Competition & Market Substitution: ANI’s primary business is news collection and syndication. ChatGPT functions as a multi-purpose tool for translation, summarizing, coding, and ideation. ChatGPT outputs do not act as market substitutes for ANI’s articles. ANI provided no evidence of lost subscriptions, diminished web traffic, or reduced ad revenues resulting from OpenAI’s operations (Authors Guild v. Google; Bartz v. Anthropic).
3. Public Interest: Training LLMs yields immense public benefits by enhancing access to information, advancing scientific research, supporting education, and driving technological innovation across vital societal sectors.

Accordingly, the Court concluded prima facie that OpenAI’s storage and processing of ANI’s literary works for training LLMs constitutes non-infringing fair dealing under Section 52(1)(a)(i).

(D) ANI Failed to Establish Irreparable Injury Warranting Interim Relief

In evaluating the final equitable principles required for interim relief, the Court held that both the balance of convenience and the risk of irreparable injury favored OpenAI:

1. ANI had previously issued a letter to OpenAI dated October 3, 2024, offering a content license for USD 7.5 million. This demonstrated that ANI’s alleged injury is fully quantifiable in monetary terms, disentitling it to an interim injunction.
2. ANI possessed the technical capability to block web crawlers and scrapers via robots.txt or paywalls on its website but chose not to do so. Furthermore, OpenAI had voluntarily implemented internal technical blocks to stop its scrapers and RAG search features from accessing ANI’s domain.
3. Granting an interim injunction would cause irreversible damage to OpenAI and millions of Indian users who rely on ChatGPT. More critically, requiring AI developers to secure advance licenses from every individual web publisher would render LLM development economically unviable which would in turn, severely hinder the growth of domestic AI innovation in India.

Parliamentary Standing Committee Hears RBI and ICAI on Virtual Digital Assets

Virtual digital assets (VDAs) are a category defined under Section 2(47A) of the Income Tax Act, 1961, covering cryptocurrencies, tokens, and NFTs. Basically, it covers any digital representation of value generated through cryptographic means. On 2 July 2026, the Lok Sabha’s Standing Committee on Finance, chaired by BJP MP Bhartruhari Mahtab, devoted a full sitting at Parliament House Annexe to “A Study on Virtual Digital Assets (VDAs) and Way Forward,” a subject the panel first took up for examination in August 2025. The sitting ran in three parts: oral evidence from the Reserve Bank of India, a session with the Institute of Chartered Accountants of India (ICAI) immediately after, and a closed-door deliberation in the afternoon. 

The RBI suggested a containment strategy tending towards prohibition. The Central Bank told the Committee that VDAs threaten financial stability and should not be granted legal status. It pointed to their use in terror financing and narcotics smuggling, the difficulty of supervising crypto held through offshore entities. Additionally, the RBI precedent of complete ban from Qatar and China, and European jurisdictions that permit it only under tight regulation. The RBI specifically raised concerns regarding stablecoins, arguing that fiat-pegged tokens undermine monetary sovereignty, and pushed instead for adoption of the central bank digital currency (CBDC). When a Committee member asked whether the RBI would instead treat VDAs as securities rather than currency, officials requested time to answer in writing. This is a question that effectively determines whether cryptocurrency will be regulated by RBI or SEBI. This is the same jurisdictional question Washington has spent the better part of a decade trying to legislate its way out of, and Mumbai appears no closer to a quick answer than Washington was.

The ICAI took a different line. It offered to build out guidance on accounting treatment, audit standards, and disclosure norms for VDA holdings, alongside research into how different VDA categories actually behave economically. ICAI’s position treats VDAs as an asset class that needs proper bookkeeping, which is a different starting position from an asset class that shouldn’t exist at all.

The hearing sits alongside a related development. Three weeks later, on 23 July 2026, the same Committee tabled its 36th Report on the Securities Markets Code, 2025. This Report noted that the Code’s technology-neutral definition of “securities” excludes VDAs that don’t independently meet the statutory test for a security or derivatives. However, many VDAs are traded and invested in the same manner as derivatives, The Committee described this as a “regulatory grey area” exposing investors to fraud and manipulation risk while inviting arbitrage. The committee proposed an interim framework built around recognised Self-Regulatory Organisations, operating under RBI or SEBI oversight, pending a dedicated statute.

That SRO-under-a-designated-regulator model is a genuinely Indian intervention. It differs from the single-rulebook approach the EU has been rolling out under the Markets in Crypto-Assets Regulation (MiCA). MiCA doesn’t split crypto oversight across institutions the way India or the US does; it puts asset-referenced and e-money tokens, service providers, and issuers all under one directly applicable regulation, uniform across all 27 member states once a firm is authorised anywhere in the bloc. The American experience offers a messier parallel, and possibly a more instructive one for India’s current moment. The CLARITY Act, attempts to split jurisdiction between the Commodity Futures Trading Commission (digital commodity spot markets) and the SEC (investment-contract assets). This is broadly similar to the division the RBI was tiptoeing around when asked about SEBI. The bill passed the House 294-134 in July 2025 and cleared the Senate Banking Committee 15-9 in May 2026, but as of yet hasn’t had a Senate floor vote. Builders have been shutting down projects over the uncertainty in the meantime. 

The numbers in front of the Committee give some sense of what’s actually at stake domestically: 54 FIU-registered VDA providers, close to 39.3 million verified users, and roughly ₹20,437 crore held, all of it already taxed at 30%, with 1% TDS on transfers since 2022. The RBI’s current posture is not new; a similar circular barring regulated entities from crypto dealings was struck down by the Supreme Court in Internet and Mobile Association of India v. Reserve Bank of India (2020) as disproportionate. In October 2025 the Madras High Court went a step further, recognising cryptocurrency as property capable of being held in trust in Rhutikumari v. Zanmai Labs Pvt. Ltd.

None of this has resolved into a single position yet. The recommendations remain advisory, and that the government may accept, modify, or decline to act on them. The Committee’s dedicated VDA study hasn’t been published yet, though a report is expected to follow soon.