Delhi High Court's AI Training Bombshell: Why ANI v. OpenAI Could Rewrite Copyright Litigation Strategy in India

The Fair Dealing Gambit That Nobody Saw Coming

The Delhi High Court has just handed the Indian technology sector its most consequential copyright development in years—and most practitioners are still digesting the implications. In the ongoing ANI v. OpenAI litigation, the Court has expressed a prima facie view that training Large Language Models on copyrighted works may fall within the ambit of Section 52(1)(a) of the Copyright Act, 1957 as fair dealing for "private or personal use, including research." On the present record, the Court suggested, such training does not amount to infringement.

Let us be clear about what this means: the single most contentious question in global AI copyright litigation—whether ingesting protected works to train models constitutes infringement—has received its first significant judicial response in India, and that response tilts decisively toward the technology companies. While this remains a prima facie observation at an interlocutory stage, any lawyer who has practised before the Delhi High Court knows that such observations often harden into the framework for final determination.

Why Section 52(1)(a) Is a Controversial Vehicle

The statutory hook here is far from obvious. Section 52(1)(a) protects fair dealing with works for private or personal use, including research. The provision was drafted in an era when "research" conjured images of scholars photocopying journal articles, not corporations scraping petabytes of data to build commercial products worth billions. The Court's willingness to entertain this interpretation suggests either an expansive reading of "research" or a recognition that India's copyright framework lacks the specific text-and-data-mining exceptions found in jurisdictions like Japan (Article 30-4 of the Japanese Copyright Act) or the EU (Directive 2019/790).

For practitioners advising AI companies or content creators, this development fundamentally alters the risk calculus. The assumption that Indian law offered no safe harbour for AI training—a position many of us have been counselling clients on—now requires urgent re-examination.

The contrast with the United States is instructive. There, the fair use doctrine under 17 U.S.C. § 107 has produced fractured outcomes: compare Authors Guild v. Google (transformative use upheld) with the ongoing NYT v. OpenAI litigation. India's fair dealing framework is narrower and more rigid than American fair use, which makes the Delhi High Court's approach all the more remarkable. The Court appears to be reading flexibility into a provision that Parliament has not substantively amended since 1994.

The DPDP Intersection Nobody Is Discussing

Here is where the analysis gets more complex. The ANI litigation does not exist in a vacuum. India's data protection architecture has undergone seismic shifts in 2026, and IP practitioners would be foolish to ignore the privacy dimensions of AI training.

MeitY's Digital Personal Data Protection (Removal of Difficulties) Order, 2026, notified on 5 October and effective from 6 October, has clarified consent requirements and—critically—revised Section 10 terminology from "audit" to "data audit" for Significant Data Fiduciaries. This is not mere semantics. The expanded audit obligations mean that any AI company training models on datasets containing personal data must now demonstrate compliance with consent frameworks that are still being operationalised.

Meanwhile, the Supreme Court has referred challenges to Section 44(3) of the DPDP Act—which gutted the RTI Act's Section 8(1)(j) personal information exemption—to a larger bench, while refusing any stay. The practical consequence: the DPDP framework continues to operate, and its consent requirements apply with full force to any entity processing personal data, including for AI training purposes.

The Court's observation in the APAAR Scheme context—that student data cannot be shared with private entities except as permitted by law—signals judicial sensitivity to unauthorised data flows. An AI company that scrapes datasets containing personal information without DPDP-compliant consent faces exposure under both copyright and privacy frameworks, even if the copyright claim itself founders on Section 52(1)(a).

Strategic Implications for Practice

For AI and technology companies: The Delhi High Court's observation provides a defensible position, but do not mistake a prima facie view for settled law. Final determination will likely turn on the specific nature of the works, the purpose of training, and whether outputs reproduce protected expression. Document your training data provenance meticulously. The "research" characterisation will face its sternest test when commercial deployment is challenged.

For content creators and rights holders: The battleground is shifting. If fair dealing provides a defence for training, the focus must move to output-side infringement—demonstrating that model responses reproduce substantial portions of protected works. This requires different evidence-gathering strategies and potentially different causes of action. Consider whether contractual restrictions on data scraping might achieve what copyright law apparently cannot.

For transactional lawyers: Representations and warranties in AI-related agreements require immediate revision. The comfortable assumption that training data licensing was mandatory now faces judicial headwinds. Conversely, the DPDP compliance burden has intensified. Due diligence must separately address copyright risk (potentially diminished) and privacy risk (potentially enhanced).

The Road Ahead

The Copyright Office's reported position that AI-generated works can receive protection while AI systems cannot be authors adds another layer of complexity. If training is fair dealing but authorship requires human creativity, we are constructing a framework where AI companies can freely ingest works but cannot claim ownership of outputs without human involvement. Whether this represents coherent policy or doctrinal confusion remains to be seen.

What is certain is that ANI v. OpenAI will be this decade's defining copyright litigation in India. The Delhi High Court has fired the opening salvo. The content industry's response—and Parliament's potential legislative intervention—will determine whether India becomes a haven for AI development or a jurisdiction where fair dealing expands beyond recognition.