Delhi Court Backs OpenAI on Training Data: What It Means for Every UK Class 42 Brand

A Third Court, a Different Road, the Same Destination

On 24 July 2026, Justice Amit Bansal of the Delhi High Court refused news agency ANI Media's application for an interim injunction against OpenAI, finding on a prima facie basis that training a large language model on ANI's copyrighted news archive falls within India's fair-dealing exception. It is the third major jurisdiction to put a judicial opinion on the record about AI training and copyrighted content and, as The New Publishing Standard noted this week, it reached the same destination through a completely different legal road.

The case, ANI Media Pvt Ltd v Open AI OpCo LLC (CS(COMM) 1028/2024), has been running since November 2024 and produced a 135-page interim order after 32 hearings. ANI, one of India's leading news agencies, alleged that OpenAI used its published news reports without permission to train ChatGPT. ANI further claimed the chatbot generated fabricated stories incorrectly attributed to the agency, a reputational-harm angle that will matter as much to brand owners as the copyright question. Justice Bansal found, on a prima facie basis, that OpenAI's storage of ANI's articles for training fell within private or personal use, including research, under Section 52(1)(a)(i) of India's Copyright Act 1957, and could constitute fair dealing. The court also found ANI had not established any memorisation or regurgitation of its works in ChatGPT's responses.

This is an interim order only. The main suit continues, and the findings do not prejudice final adjudication.

Why It Matters Beyond India

For UK brand owners and their IP advisers, the geopolitical spread of this case is the point. Indian copyright law's Section 52 operates as a closed statutory list, not the flexible multi-factor fair-use test used in US courts. The Delhi bench read the word "research" in that list against present-day AI technology rather than its 1957 meaning. That interpretive move is now on the record. OpenAI and every other lab training on scraped news will cite it in parallel suits working through US and Canadian courts.

The European Commission's IP Helpdesk flagged the ruling on 28 July, noting that ANI had brought proceedings on two bases: use of its content during training, and reproduction of its articles in AI-generated outputs. The court dismissed both at the interim stage. The attribution-and-hallucination question, ANI's claim that ChatGPT generated false stories bearing its name, was found not to establish a prima facie infringement case either. That second limb, less commented on, is arguably the more acute brand-protection issue: it is one thing for a court to permit training on your archive; it is another to permit a system to generate fabricated content under your registered marks.

World Trademark Review's homepage this week summarised the Delhi ruling as boosting the AI training defence while leaving copyright battles far from settled. That caveat is well-placed. Justice Bansal also rejected OpenAI's jurisdictional challenge, confirming that Indian courts may hear claims about AI training conducted on US servers if the effects are felt in India. That ruling on jurisdiction cuts both ways: it keeps future suits inside courts that may rule differently at trial.

The Class 42 Signal in the UK Filing Data

This ruling lands at a sensitive moment for UK technology brands. According to AIBD analysis of IPO data, UK trademark filings in Class 42 (the class covering scientific and technological services, software-as-a-service, and, since 1 January 2026 under the Nice Classification's 13th edition, Artificial Intelligence as a Service (AIaaS) explicitly) reached only 2,635 in Q3 2026, a drop of 64.4% against the prior period (AIBD analysis of IPO (TMD) data, as of July 2026). That contraction is striking. Class 42 is precisely where AI service brands need protection, and the filing decline suggests either strategic hesitation or a misreading of the landscape.

The 13th edition of the Nice Classification, in force from 1 January 2026, formally added AIaaS as a recognised service in Class 42, giving applicants a precise, internationally harmonised term for AI service delivery. Any UK business providing AI-powered services that has not yet filed in Class 42 using that express language is leaving a gap in its portfolio, one that the Delhi ruling does nothing to close. Copyright and trademark protection are distinct instruments. A court finding that your training data is lawful to use does not protect the brand name under which you deploy the resulting model.

There is a second gap to watch. ANI's hallucination claim, that ChatGPT published fabricated content attributed to the agency's registered brand, points to an emerging category of AI-related passing off and trademark misuse that no court has yet settled on the merits. UK brand owners in media, data and professional services should be auditing whether their registered marks appear in AI outputs in ways that suggest affiliation or origin they have not authorised.

What to Do This Week

The Delhi ruling is interim, not final. But its logic, that commercially operated AI labs can claim research-purpose fair dealing, is now persuasive authority in a significant jurisdiction and will be cited widely. The correct response for UK brands is not to relax; it is to recognise that copyright protection and trademark protection must work together.

If you have not filed in Class 42 using the AIaaS term introduced in Nice 13, consult a trade mark attorney or run a preliminary portfolio check via TrademarkDashboard to identify the gap. The UK IPO keeps your £205 filing fee if an application is refused, so clearance before filing remains the cheaper option. If your brand name or registered marks appear in AI-generated content you have not authorised, document those instances now. The hallucination-and-attribution question will come before a court that does rule on the merits, somewhere, before long.