By: Aashrith Ramineni

INTRODUCTION: A GROWING THREAT IN THE AGE OF SYNTHETIC SPEECH

AI-driven voice cloning is transforming how content is produced, enabling creators like Varun Mayya to automate their entire video production process using a digital replica of their own voice. With just a few seconds of recorded audio, these tools can generate lifelike speech at scale, streamlining workflows and making mass content generation easier than ever. However, the same technology is increasingly being misused for fraud, impersonation, and other malicious purposes, highlighting the dual-edged nature of its impact. A recent study found that 47% of Indians have personally experienced or know someone who has suffered a voice-cloning scam. In Lucknow, a young man sent ₹45,000 to what he thought was a desperate plea from a relative – only to discover the relative was an AI-generated impostor. Similar scams have hit victims nationwide using emotionally manipulative fake voice calls. Abroad, too, fake voices are weaponised. In Sudan’s conflict, deepfake audio allegedly impersonated political figures (like ex-president Omar al-Bashir) on social media, sowing panic and disinformation. These incidents show that synthetic voices threaten broader informational trust and security, because, making scams, defamation, and political manipulation cheap and hard to trace.

This article argues  how India’s current reliance on copyright and personality rights leaves serious gaps in the AI voice-cloning age, and argues for a harms-based approach that combines post-harm remedies for victims with pre-harm safeguards like labelling, watermarking, and rapid takedowns, drawing lessons from global models on how to implement these protections in practice.

GAPS IN COPYRIGHT LAW AND PERSONALITY RIGHTS

Under the Copyright Act, protection is tied to a ‘work’ that is fixed in some form, and Section 17 mainly clarifies that the ‘author’ is the first owner of copyright. A person’s voice, by itself, is not treated as a protected work unless it is embodied in a recorded dramatic, musical, or artistic performance. Even though the Act recognises a public speech or address as a ‘work’ whose copyright belongs to the speaker, that protection remains limited to fixed, public utterances, meaning it covers a recorded performance, not the mere sound or qualities of someone’s voice. As a result, an artist’s voice is not inherently protected unless it exists within a recording. This is consistent with US jurisprudence as well: in Midler v. Ford, the court held that a distinctive voice alone is not fixed and therefore not copyrightable. In India, performers’ rights under Section 38A offer protection for 50 years, including consent and royalty requirements, while Section 57 provides moral rights against distortion or mutilation. Yet cases such as Indian Performing Right Society Ltd. v. Aditya Pandey and MRF Ltd. v. Metro Tyres Ltd. show that fixationremains central, leaving limited clarity for AI-generated uses. Since AI systems often generate new speech based on samples that mimic a voice rather than copying a specific recording, many voice clones escape traditional copyright categories.

In contrast, personality rights are judge-made protections that prevent unauthorised use of a person’s identity traits such as their name, image, likeness, or voice without consent. Because there is no statute, Indian courts have developed these rights through case law, and they often frame them in property-like terms: something a person can commercially control, license, and monetise. This approach has meant that voice is frequently treated less as an aspect of dignity, autonomy, or selfhood, and more as a marketable asset with economic value. This commercial framing has shaped judicial responses to voice misuse, especially in the AI context. Courts have increasingly extended personality rights to AI-generated voice clones, but mostly in celebrity cases where brand value and commercial harm are easier to establish. For instance, in Arijit Singh v. Codible Ventures (2024), the Bombay High Court restrained unauthorised AI replication of Singh’s voice, holding that his voice was protected under personality rights. Similarly, Amitabh Bachchan v. Nagi (2022) and Anil Kapoor v. Simply Life (2023) recognised a celebrity’s face, name, and voice as monetisable elements of identity.

However, this personality-rights model is not well suited to the AI voice-cloning era. Because courts often require proof of commercial value or brand identity, protection tends to be strongest for public figures, while ordinary individuals struggle to invoke the same level of legal protection. This creates a practical two-tier system: celebrities receive robust remedies, but non-celebrities remain far more vulnerable to cloning and misuse.

Statutory law offers little support because copyright law continues to treat voice, if at all, as part of a copyrighted work rather than an inherent feature of identity. Many AI voice clones therefore fall outside classic copyright categories such as reproduction or protection for fixed public speeches, making copyright remedies uncertain. Therefore, copyright law and the current personality-rights framework are still poorly equipped to protect the wider public from AI voice cloning. The deeper harms of voice cloning, including violations of dignity, autonomy, and trust, remain largely unaddressed for most people. Gurumurthy’s analysis of Indian case law on deepfakes highlights how courts have often failed to acknowledge these layered harms. This shows why India needs a clearer harms-based legal approach that treats voice cloning as a direct violation of dignity and autonomy.

WHY INDIA NEEDS A HARMS-BASED FRAMEWORK AND WHAT GLOBAL MODELS TEACH US ABOUT IMPLEMENTATION

India’s current approach leans too heavily on personality rights, which leaves most victims without meaningful remedies and ignores wider harms like disinformation, privacy invasion, emotional injury, and the erosion of public trust. A harms-based approach would shift the focus from who the victim is to what the cloning does. Legal protection should turn on whether a cloned voice causes fraud, reputational damage, emotional distress, political manipulation, or public confusion, not on whether it appears in an advertisement or a film. This shift would extend remedies to everyone, including children, private citizens, and public figures whose voices are misleadingly used, and it would also empower regulators to respond to social harms like deepfake extortion and fake news.

Reform, therefore, must be two-fold: post-harm remedies that give victims enforceable rights once misuse occurs, and pre-harm safeguards that reduce the spread and impact of synthetic media before it escalates. On the post-harm side, copyright and personality rights are still only partial fits, so reforms could also introduce sui generis protections. Sui generis means a standalone, tailor-made legal right (‘of its own kind’) created for a new category of harm that existing doctrines cannot fully address. In this context, a sui generis ‘voice and likeness’ law would recognise a person’s voice as a protected attribute and set out clear rules on consent, liability (covering creators, uploaders, and in some cases platforms), and remedies such as damages and injunctions.

At the same time, any new regime must be technically implementable, which is where pre-harm safeguards become essential. The draft Synthetic Media rules proposed by India’s IT Ministry show the way. They would mandate watermarking or labelling of AI-generated content so it is clearly flagged as synthetic. For example, visible tags on video or brief audible disclaimers in audio (covering at least 10% of content) could alert platforms and users.  AI developers would be required to embed imperceptible metadata or signatures in their models, enabling detection tools to spot fakes. Upstream accountability is key: similar to the draft rule, upstream ‘tool providers’ (companies offering AI voice or video generators) should be legally obligated to mark every output and to keep logs tying clones back to source models.

On the platform side, social media and content hosts must act quickly on reports of harmful clones.  Existing intermediary rules (Section 79 of the IT Act) already protect platforms when they remove illegal content in good faith.  This can be extended to fast-track synthetic media.  The MeitY draft mandates that large intermediaries obtain user declarations about synthetic content, use automated checks, and take down flagged fakes within hours. Judges have also begun demanding technological solutions: in the recent Sadhguru deepfake case, the Delhi High Court ordered YouTube to deploy automated filters to catch and delete infringing clones within 36 hours.  India can build on this by requiring platforms to use watermark-detection tools and by empowering regulators (or courts) to block accounts that churn out dangerous deepfakes.

Global experience shows these measures are practical.  The European Union is making progress: the forthcoming AI Act (Article 50) will oblige providers of generative AI to mark all outputs so they are machine-readable as synthetic , and deployers (e.g. social media firms) to disclose AI-generated ‘deep fakes’ that mimic real people.  Likewise, China’s new deep synthesis rules require that any AI-generated content carry a ‘noticeable mark’ indicating it is synthetic. India’s law could tie compliance to these technical rules for example, making unlabeled deepfakes subject to swift takedown or fine.

The most workable harms-based response must remain two-fold. Post-harm protections (especially sui generis rights) will take the most legislative effort because they require a new statutory framework for consent, liability, and remedies. Pre-harm safeguards are where India is already moving in the right direction through draft watermarking, detection, and takedown duties, but the system still needs more concrete, enforceable standards and timelines to prevent synthetic media harms from spreading faster than the law can respond.

BALANCING HARMS AND FREE SPEECH

A common concern is that strict AI regulation could chill free speech under Article 19(1)(a), especially satire, parody, and political expression that uses synthetic media. Any new framework must therefore protect legitimate speech. Denmark’s proposal, which recognises rights over one’s appearance and voice (sui-generis right), explicitly exempts parody and satire. The EU follows a similarly balanced approach: AI content used in news or political commentary may not require a separate ‘AI-generated’ label if it has undergone human editorial review, since responsibility is anchored in editorial accountability.

India should adopt comparable safeguards. Public-interest satire and fair reporting using voice synthesis should be presumptively permitted. Regulation should instead focus on harmful misuse, particularly non-consensual cloning, and even consensual cloning where the output is deployed deceptively to cause fraud, impersonation, misinformation, or public confusion. This is consistent with a harms-based model: liability should turn on the real-world effects of voice cloning such as reputational damage, emotional distress, deception, or loss of trust rather than the victim’s fame or whether the use is commercial. This approach targets harm without suppressing legitimate expression.

CONCLUSION

AI voice cloning threatens far more than celebrity privacy or copyright. It enables fraud, reputational damage, emotional injury, and large-scale distrust, while ordinary people often have limited  legal protection. Indian courts have responded through personality-rights cases, but this remains celebrity-focused and does not capture wider harms.

India is seriously lacking in post-harm remedies: victims still lack clear statutory rights, predictable liability rules, and effective relief when cloning causes real injury. That said, India has taken minor steps in the right direction on pre-harm regulation through draft proposals on watermarking, disclosure, and faster takedown systems. But these steps need enforceable standards and timelines. A future-ready framework must strengthen both post-harm protections (including possible sui generis rights) and pre-harm safeguards, while keeping speech carve-outs for satire, parody, and public-interest uses. Only such a harm-oriented, balanced approach can safeguard citizens’ dignity and maintain trust in communication in the digital age.

(Aashrith Ramineni is a third-year B.A. LL.B. (Hons.) student at National Academy of Legal Studies and Research, Hyderabad. The authors may be contacted via mail.)