Intellectual Property Enforcement in the Epoch of Generative AI

The Training Data Crisis: Infringement vs. Fair Dealing

The ingestion of massive, copyrighted datasets to train Large Language Models (LLMs) and latent diffusion models (like Midjourney or Stable Diffusion) has triggered global litigation. In India, the debate centers on whether such scraping constitutes mass infringement under Section 51 of the Copyright Act, 1957, or falls within the protective ambit of 'fair dealing' under Section 52.

Section 52 allows for limited use of copyrighted material for purposes such as private research, criticism, or review. However, AI developers utilize works not for human 'consumption' but for algorithmic pattern extraction—a process fundamentally commercial in nature. While global courts (e.g., 'Getty Images v. Stability AI' in the UK/US) grapple with this, Indian jurisprudence is preparing for landmark clashes. Until a definitive Supreme Court ruling, copyright holders must rely on technological protection measures and explicit anti-scraping clauses in their Terms of Service.

The Authorship Dilemma: Section 2(d)(vi)

Can an AI be an author? Section 2(d) of the Copyright Act defines an 'author' strictly in human terms. However, Section 2(d)(vi) specifically addresses computer-generated works, stating that the author is the 'person who causes the work to be created.'

The legal complexity arises when an AI generates a sophisticated output from a simple text prompt. Is the prompter the author, or merely the 'commissioner' of the work from the AI? The Indian Copyright Office issued a withdrawal notice for a registration previously granted to an AI app (RAGHAV) as a co-author, signaling administrative resistance to non-human authorship. Courts are likely to adopt a 'sweat of the brow' and 'modicum of creativity' test (as laid down in 'Eastern Book Company v. D.B. Modak (2008) 1 SCC 1') to determine if the human prompter contributed sufficient original expression.

Personality Rights and Dynamic Injunctions

Perhaps the most immediate threat posed by Generative AI is the proliferation of deepfakes and unauthorized voice cloning, severely impacting celebrities and brand ambassadors. India lacks a specific statute for personality rights, which are instead enforced as a subset of the right to privacy (Article 21) and passing off under Trademark law.

The Delhi High Court has been remarkably proactive in this domain. In the landmark case of 'Anil Kapoor v. Simply Life India & Ors. (2023)', the Court granted a sweeping ex-parte omnibus injunction protecting the actor's name, likeness, voice, and even his catchphrases from unauthorized AI-generated commercial exploitation. This established a robust precedent for securing dynamic injunctions against unknown defendants ('John Doe' orders) to combat the rapid digital dissemination of AI deepfakes.

Strategic IP Management in 2026

For IP practitioners and corporations, the strategy must shift from reactive to proactive. Trademark portfolios must be expanded to cover digital and virtual manifestations. Entertainment contracts must now explicitly address and ring-fence the rights to digital replication, voice synthesis, and AI training data licensing.