When the Delhi High Court’s ruling in ANI v OpenAI dropped on a Friday evening, many of us were genuinely excited. It is India's first real answer to whether training an AI model on copyrighted content amounts to infringement. This edition unpacks this judgment, digging into what ANI argued, and how the Court worked through the questions before it.
The Court did not restrain OpenAI from using ANI's articles to train ChatGPT while the case continues. Although it found that storing copyrighted works for AI training would ordinarily amount to reproduction under the Copyright Act, 1957, it held that OpenAI's use was protected by the Act's fair dealing exception at this stage. The trial is still ahead, but for now, OpenAI has the upper hand—and we explain why in this month's deep dive.
In the rest of the edition, we track how government intervention in digital platforms appears to be moving beyond individual pieces of content and towards platform architecture, and executive accountability.
I. I. Deep Dive
India's first AI and copyright ruling
India finally has its first ruling on AI model training and copyrighted content. If you don't have the time to read the 135-page judgment and would rather spend that time watching Nolan’s Odyssey or Spiderman instead, don’t worry. We’ve got you covered.
Here are the questions you probably have about the case (and if we've missed one, write to us).
A A. Who sued OpenAI, and why?
In November 2024, Asian News International (ANI), an Indian news agency, sued OpenAI before the Delhi High Court (Court). ANI claimed it owns the copyright in its work - mostly news articles (which we will refer to as "content"). As the owner, it said, it has the exclusive right to reproduce or store that content.
ANI alleged OpenAI used and stored its content, without permission or a license, to train ChatGPT’s underlying LLM. It also alleged ChatGPT sometimes falsely attributed statements to ANI that it never made. ANI asked the Court to stop OpenAI from accessing, storing, or reproducing its content to train its model.
B. What key questions did the Court look at?
At the first hearing, the Court framed four questions to decide the case:
- Can Indian courts hear this case, given that Open AI’s servers are in the US (Jurisdictional Issue);
- Does ChatGPT infringe ANI’s copyright when it uses ANI’s data to answer user queries? (Output claim)
- Does storing ANI’s works to train ChatGPT infringe its copyright (Training Claim); and
- Can OpenAI claim the fair dealing exemption under the Copyright Act, 1957 (Act)? (the fair dealing question)
C. Can Indian courts hear this case? (Jurisdiction Issue)
Yes, they can.
ANI is based in New Delhi, giving the Court jurisdiction under Section 62(2) of the Copyright Act. The Court noted that OpenAI’s US-based servers don’t change this. ChatGPT specifically targets users in India, and the training data collection took place from India. Even the responses generated are based on prompts entered in New Delhi. So, the case has an Indian nexus regardless of where the data ends up being stored.
D. Did ChatGPT’s responses infringe ANI’s copyright? (Output Claim)
No, not at this stage.
The Court reaffirmed that ANI holds copyright in its news articles, and that making them publicly available does not dilute those rights. But it found that ChatGPT's responses, that ANI relied on in its pleadings, don’t substantially reproduce ANI’s content, which is the threshold for establishing infringement. At this stage, ANI had failed to make out a case on the Output Claim. So, there is no infringement.
Some questions are still left open to trial – where both parties will present evidence and witnesses to determine the issues in finality. These include whether ChatGPT permanently stores copyrighted material during training, memorises it, and later reproduces it in its responses.
E. Does storing ANI’s works to train ChatGPT infringe its copyright (Training Claim)?
No – because of the fair dealing exemption.
Since ANI holds copyright, it also holds the exclusive right to reproduce them. This right of reproduction includes the right to store the work “in any medium by electronic means.” The Court noted that this scope is deliberately wide. So storing a copyrighted work, even for LLMs, counts as reproduction. The law doesn’t distinguish between temporary and permanent storage under Section 14, and the intention behind the storage is irrelevant for determining infringement.
So storing ANI’s articles to train ChatGPT counts as reproduction — a right that belongs exclusively to ANI under Section 14. For this reproduction, OpenAI would normally need a licence.
But the Court emphasised that these exclusive rights are not absolute. They are subject to the exceptions in Section 52 of the Copyright Act, including the fair dealing exception for "private or personal use, including research." The real question, therefore, was whether OpenAI's use fell within that exception.
The Court answered this in the affirmative. It held that OpenAI stores the articles in a closed environment that is inaccessible to the public and uses them solely for training its large language model, making the use "private." It also found that the process of collecting, organising, analysing and converting the material into machine-readable inputs for model training qualifies as "research."
The Court also found the use to be fair because it was limited to training the LLM, did not prejudice ANI's legitimate commercial interests, and served the broader public interest. It therefore held that OpenAI's use fell within the fair dealing exception under Section 52 and did not require ANI's permission.
F) F. What happens next?
We expect ANI to appeal. This ruling gives the first sense of how Indian courts may treat AI training under the Act, and will likely shape future cases, giving AI companies operating in India a similar defence to rely upon. But some findings are fact-specific, and a case presented differently before the Court may lead to a different outcome.
It’s possible that if the works in question were not news articles, which carry weaker copyright protection, but other works like music or books, which have a higher threshold of creativity, the outcome could be different. The jurisdiction finding also means Indian rights holders can now sue foreign AI companies in India, ultimately inviting more such cases before Indian Courts.
We have a more detailed analysis of the case in our post here.
II. Connecting the dots
After Telegram, are peer-to-peer messaging apps next to be banned?
Last month, we flagged that the Delhi High Court's order upholding the Section 69A ban on Telegram could set a precedent for the government to block an entire platform, given the right circumstances. This month, we saw an example of that taking potentially taking place.
MeitY is already considering common rules for the username feature, after sending notices to WhatsApp, Telegram and Signal asking them to explain safeguards around letting users message without revealing their phone numbers.
During the ongoing Jantar Mantar protests, the Indian Cyber Crime Coordination Centre (I4C, the Ministry of Home Affairs' nodal cybercrime body) reportedly sent separate takedown notices to Google, Apple and GitHub, each targeting peer-to-peer messaging apps such as BitChat, Briar and Bridgefy.
All notices demanded compliance within three hours, citing Section 79(3)(b) of the Information Technology Act and Rule 3(1)(d) of the Information Technology Rules. I4C flagged the apps' ability to function without internet, centralised servers or registration as a threat to lawful interception. This was an argument like the platform architecture argument made during the Telegram ban. Within 24 hours, the government orally told all three companies not to comply. There was no implication on the ground, no app left either store. GitHub held off pending owner notification, and BitChat's developer Jack Dorsey tweeted the notice publicly.
Whether or not these notices ultimately resulted in any enforcement, they reinforce an emerging trend. The government's attention appears to be shifting beyond individual pieces of content towards the design of digital platforms themselves. Platform architecture and the possibilities it creates is becoming a subject of regulatory scrutiny.
Safe harbour gets personal
On 3 August, the SCIT (Parliamentary Standing Committee on Communications and IT) held a three-hour hearing with executives from Meta, Google/YouTube, X and Snap. The immediate trigger was Meta's brief removal of a PM Modi video, which Meta's India public policy head apologised for as an algorithmic error. Committee chair Nishikant Dubey rejected the explanation, demanding a personal apology from CEO Mark Zuckerberg. More significantly, he recommended that the IT Secretary consider withdrawing platforms' safe harbour under Section 79 for non-compliance and reclassifying social media platforms as publishers rather than intermediaries.
The hearing surfaced compliance fights beyond the Modi video too. The MHA and I4C pressed platforms on their refusal to report CSEAM (child sexual exploitative and abuse material) to local police, citing a 2024 Supreme Court ruling requiring dual reporting to NCMEC and India's Special Juvenile Police Units.
The pressure extends beyond corporate liability. Hyderabad Cyber Crime Police have registered two FIRs against Meta India MD Arun Srinivas over posts targeting PM Modi, while the Central government has separately met Meta's global team over CSAM concerns, continuing last month's scrutiny of Meta's content moderation and advertising systems.
During a meeting between MeitY and Meta's senior global executives, Meta acknowledged errors in operating its platforms and reportedly admitting that money had been paid to boost certain categories of content.
Regulators are looking beyond the platform as an intermediary and towards the executives responsible for its compliance. Safe harbour may continue to shield platforms from liability for user-generated content, but it appears to offer far less protection from regulatory scrutiny of their leadership.
III. From the courtroom to your inbox
- Delhi High Court asked to bar under-13s from social media: On 5 August, the Delhi High Court heard a PIL seeking a nationwide framework restricting social media access for children under 13. The petition also asks for content controls for 13–16 year-olds, overnight curfews for teen users, and mandatory age verification modelled on Digi Yatra. Alphabet, Meta, Telegram, Snap, and X are named respondents. During the hearing, the Court noted that a previous, similar petition, Zep Foundation v. Union of India, had already been dismissed by the Supreme Court, with the petitioners there permitted to make a representation to the federal government on these issues. It directed the central government to seek instructions on what’s happened to that representation since and listed the matter next for 20 August.
- Supreme Court imposes INR 3 lakh fine on YouTuber Samay Raina and four others for defying its orders: The Supreme Court imposed a fine of INR 3 lakh each on YouTuber-comedian Samay Raina and four others for violating its November 2025 order directing them to host at least two fundraising events a month for a disability-treatment corpus. The Court found Raina had not involved persons with disabilities in his show as required. The proceedings trace back to a Cure SMA India Foundation petition over remarks made on the show about the cost of treating Spinal Muscular Atrophy, and the petition separately seeks regulations on the broadcast of such content, invoking the right to life and dignity of persons with disabilities.
- A PIL to discuss Instagram's copyright strike button: A digital content creator filed a PIL before the Delhi High Court alleging an organised racket that exploits Instagram's automated copyright enforcement. Criminal groups file fake copyright complaints to get legitimate accounts suspended, then demand payment to restore them. The PIL seeks human review before account suspension and a time-bound appeals process. Another similar matter has been filed by financial educator Pushkar Raj Thakur and creator Neeraj Joshi before the same court on 16 July. The court directed Meta to preserve Joshi's account and share IP logs; Meta assured it won't remove Thakur's videos or close his account before the next hearing. It will be heard on 21 September next.
IV. Reading reccos
- Ben Thompson in Stratechery explores explores the debate around open-weight models and the scrutiny of Chinese LLMs.
- The founder of Patreon Jack Conte has a compelling presentation on the future of being a creator in a world with AI.
- The BBC has a good report why the economics of agentic AI remain challenging, with token costs making sustainable business models difficult to build
V. Shoutouts
- Nehaa has been reappointed as the Co-Chair of the ITechLaw Association India Conference.
- Aparajita and Astha conducted a webinar for the FinTech Association for Consumer Empowerment's (FACE) knowledge initiative on "India's KYC Framework: Recent Shifts and the Road Ahead," unpacking recent regulatory developments and what they mean for the future of KYC compliance.
- Ikigai Law supported the Startup Policy Forum at Semicon Baatcheet, chaired by Shri S. Krishnan, Secretary, MeitY. The roundtable convened government, startups, investors and industry leaders to discuss practical recommendations for implementing India Semiconductor Mission (ISM) 2.0 and strengthening India's semiconductor ecosystem.
Signing off
Ticker team for this edition:
