Authored By: Shivam Bansal
Bhai Gurdas College of Law, Sangrur
INTRODUCTION TO AI VOICE CLONING
Earlier this year, a group of actors in the UK launched the “Save Our Voices Now” campaign, pushing for legal protection against having their voices cloned and reused without consent. That’s telling in itself. These are people with agents, lawyers, and enough public profile to make noise when something goes wrong and they’re still worried.
A few years ago, cloning someone’s voice took a film studio’s budget and a sound engineer who knew what they were doing. Now it takes about three seconds of audio and a free app. AI voice cloning, training a generative model to reproduce a real person’s voice from a short sample, has gone from a research curiosity to something anyone can do on a phone, often without the original speaker ever finding out. If performers with legal resources and public visibility are organising against this, someone without any of that, whose voice has no commercial value anyone would think to protect, is arguably far more exposed, not less.
MISUSE OF VOICE CLONING
None of this stays theoretical for long. Once a cloned voice sounds close enough to the real thing, the danger stops being about money and starts reaching into almost everything a voice is used for — trust, identity, even proof that something was actually said. A machine that sounds like a human, knows something about you, speaks in your language, responds to your questions and imitates the voice of someone you know can be turned into a powerful instrument of surveillance, manipulation, impersonation and commercial intrusion. A recent study found that 47% of Indians have personally experienced or know someone who has suffered a voice-cloning money scam[1]. Cloned voices have also surfaced in non-consensual sexual content, fabricated defamatory statements attributed to public figures, and manufactured audio “evidence” in personal or legal disputes, where a fabricated recording can be nearly impossible to distinguish from genuine one without forensic analysis.
SYNTHETIC IMPERSONATION
Each of these harms, on its face, looks like something Indian criminal law already punishes- cheating, impersonation, forgery, defamation. Whether it actually does depends on a distinction the law has never had to draw before: the difference between impersonation committed by a person, and impersonation committed through a person’s own identity, without them being involved in the act at all. The distinction is what the term “synthetic impersonation” is meant to capture.
What makes voice cloning legally distinct from every other form of impersonation the law has dealt with so far is where the act of deception actually happens. Traditional impersonation, whether under criminal law or common understanding, requires a person to do something i.e., either speak in another’s name, or forge a signature, or adopt a false identity in conduct or communication. The impersonator is present in the act. Voice cloning removes that presence entirely. Once a model has been trained on a few seconds of someone’s speech, the fraud is committed in that person’s own voice, without them saying a word, without their knowledge, and often without them ever finding out until after the harm is done. It does not target a single victim’s existing identity. It builds an entirely new one.
What makes a clones voice such a difficult instrument to guard against is that it is built from a mixture of the real and the fabricated. The voice itself is genuine, drawn directly from the actual person’s recorded speech, carrying their real vocal bio-metrics, tone and cadence. What’s fabricated is everything the voice is made to say. This is why “synthetic impersonation” is doing real work as a term rather than simply being a label. It marks a category of harm where the victim’s own biometric identity becomes the instrument of the crime committed against them or through them, all without the actual person being involved at any stage beyond having once spoken in a way that got recorded.
LIABILITY UNDER BHARATIYA NYAYA SANHITA,2023
The non-involvement of the person whose identity is used in voice-cloning is precisely the trap facing existing law. Offenses like cheating, impersonation, and forgery were drafted around the assumption that a person performs the deceptive act. Synthetic impersonation breaks that assumption at its foundation: there is no impersonator in the traditional sense, only a model trained to output a voice that isn’t lying, but it isn’t speaking at all, it’s generating. The question is whether the provisions built for a human actor can stretch to cover an act with no human actor at the point of impersonation.
If a cloned voice is used to defraud someone in India today, the first place a ploce officer would look is the Bharatiya Nyaya Sanhita,2023. The provisions of BNS like cheating, cheating by personation, and forgery can each, with some effort, be pointed at voice cloning cases but none of them fits cleanly.
Cheating, under section 318 of BNS, involves deceiving people to induce them to suffer wrongful loss through giving up money or making them to do anything which would cause damage to their body, mind, reputation or property. It also includes giving of confidential information like, bank account details or one time passwords etc. Therefore, a scammer using a cloned voice to convince someone’s parent that their child is in danger and needs money fits fairly well as there is deception, there’s inducement and there’s a dishonest gain. And hence, in this way the provisions related to cheating are applicable in voice cloning cases and is probably the strongest fit in the entire BNS for financial voice-cloning scams.
Section 319, cheating by personation, is where things get tricky. This provision punishes someone who cheats by pretending to be another person, or by representing themselves as someone they know they aren’t. On paper, this looks perfect fit for voice-cloning. But in practice, the word “pretending” carries an assumption, i.e., a person is doing the pretending. When a scammer clones a father’s voice to call his son, who exactly is “pretending” to be the father? The scammer himself does not do anything, he never speaks in the father’s voice. He just scans the original voice sample and types a script and then, the AI model comes into motion. It itself generates and does the impersonation. The human being who committed the fraud never personated anyone, a piece of software did the actual impersonating. The human just gave the instructions. Courts haven’t yet decided that whether directing a machine to impersonate someone counts as “personation” or not. Until the question is settled, section 319 remains a plausible charge, not a guaranteed one.
Forgery provisions also don’t provide any substantive ground. The forgery provisions are built around the concept of a false document, something written, signed, or otherwise recorded to appear as though it came from someone who didn’t actually create it. A cloned audio file, here, sits in an uncertain territory. It is unclear whether a synthetic voice recording can deemed to be a document for the purpose of bringing it under the ambit of forgery. At present, no Indian judgment has answered this directly.
There’s also a mens rea problem that runs underneath all three provisions. Cheating and personation both require proof of dishonest intention i.e., the prosecution has to show that the accused meant to deceive. That’s usually straightforward when a human being personally speaks the lie. It gets tricky with voice cloning, because the “deception” is generated by a model, and the human being behind it might argue they only built or used a tool, without directly authoring each fraudulent statement the tool went on to produce. This is a thin argument, and courts would likely see through it in most cases but it’s exactly the kind of gap a skilled defence would try to open. It points to something deeper in how Indian criminal law decides who did what. Before asking whether any of these provisions technically apply, it’s worth asking a more basic question first: when a machine generates the deceptive act on a person’s instruction, who, in law, actually committed it?
ATTRIBUTION OF CONDUCT
The main difficulty with AI voice-cloning lies in tracing the fabricated elements back to a specific person. Indian criminal law, like most legal systems of the world, attributes conduct on the person whom committed the act. The person forms an intention, that intention produces an act, and the act causes the harm. But the phenomenon of voice-cloning breaks this chain. Firstly, because the “act” itself is no longer performed by the accused. It’s performed by a model, on the accused’s instruction. Thus, there is no person to trace back the act to, just a computer model.
Secondly, one could argue that the “act” was performed by the software, and the human being’s conduct is closer to abetment or conspiracy than to direct commission of the offence. But the Indian law was built with human beings in mind and not machines. Abetment under the BNS punishes someone who instigates, conspires with, or intentionally aids another person in committing an offence. The entire framework presumes two persons: the principal offender and the abettor. Voice-cloning software doesnot fit either role. It isn’t a person who can be instigated or who instigates. It has no intention of its own, no capacity to be a offender in the legal sense.
But there is a temporary solution to this problem. The courts have long treated a weapon, a forged seal as an instrument of the offence rather than a competing actor and on the basis of this analogy, a voice-cloning model I simply a more sophisticated instrument and the person who directs it should be treated as having committed the act, exactly as if they’d spoken the words in disguise.
WHO IS RESPONSIBLE?
Moreover, these crimes often arise from long technological chains and not from one lonely wrongdoer. There’s often the person who built or trained the model, the person who fed it a target voice and generated the clip, and, separately again, the person who put that clip to use i.e., played it down a phone line, posted it online, or forwarded it to someone who’d believe it. Model creators, platform providers, app developers, prompt engineers and end users all participate at different stages of harmful content lifecycle.
Once you accept that a cloned voice usually passes through more than one set of hands before it does any damage, a harder question follows: which of those hands should the law actually reach for? Traditional offences in Bharatiya Nyaya Sanhita still expect clear mental element in one identified accused. Treating all these as equally culpable feels intuitively wrong, and Indian law, as it stands, has no clean way of sorting them. But when one person codes model, another hosts service, third uploads prompt and hundreds amplify content, locating mens rea becomes doctrinally messy.
The honest answer is that Indian law currently pulls all three of these actors, the builder, the user, and the sharer, through the same narrow set of provisions built for cheating and forgery, provisions that were never designed to distinguish between degrees of involvement in a multi-step technological process. A statute that actually addressed synthetic impersonation would need to draw these lines explicitly: heavier liability for whoever generates a clone with intent to deceive or harm, some liability for platforms or developers who knowingly enable clearly illegitimate use, and a separate, lighter framework for people who share synthetic content without full knowledge of its origin or purpose. Right now, none of that line-drawing exists in Indian statute.
PROPOSED LEAGL FRAMEWORK
None of this means Indian law has done nothing about voice cloning. Cheating, personation, and forgery can all be pushed towards a case with enough effort. But the reality is that none of these tools were built for this specific problem, and stretching them case by case leaves too much resting on how one judge happens to read one word.
There are early signs of movement. Reports of a draft Synthetic Media Bill have circulated through 2026, aimed at giving individuals clearer rights over their own voice, with penalties attached to unauthorised replication done for impersonation or harm. It remains at the draft stage — nothing has reached Parliament yet, so this is a direction, not a protection anyone can currently rely on. MeitY has also floated advisories on labelling AI-generated content, alongside years of talk about replacing the IT Act with something built for AI-specific harms.
The clearer fix is a dedicated offence, criminalising the non-consensual cloning of a voice for deception or harm, separate from cheating and forgery, so the law stops needing a human being to have personally “impersonated” anyone. Liability should scale with proximity: harsher penalties for whoever generates a clone with intent to deceive, lighter but real liability for platforms that knowingly enable misuse, and a civil track for people who share synthetic content without knowing its origin.
CONCLUSION
Taken together, the problem of AI voice-cloning doesnot require India to invent an entirely new area of law. It requires acknowledging what this paper has tried to show throughout that cheating, personation, and forgery were built around a person doing the deceiving, and voice cloning was never going to fit that mould no matter how creatively the provisions get interpreted. Until then, Indian criminal law will keep doing what it’s doing now: reaching for provisions that were never written for this, and hoping the analogy holds up in court.
REFERENCE(S):
Legislations
Bharatiya Nyaya Sanhita, 2023
Information Technology Act, 2000
Secondary Sources
Aashrith Ramineni, ‘AI Voice Cloning in India: Rethinking Legal Frameworks Through a Harms-Based Lens’, (The RMLNLU Law Review Blog, 25 June 2026) https://rmlnlulawreview.com/2026/06/25/ai-voice-cloning-in-india-rethinking-legal-frameworks-through-a-harms-based-lens/ accessed 1 September 2026
Dr. Neeraj Jain, ‘AI Voice Cloning Frauds: Emerging Challenges for Indian Criminal Law’, (Indian Journal of Law and Legal Research, 8 August 2026), https://www.ijllr.com/post/ai-voice-cloning-frauds-emerging-challenges-for-indian-criminal-law accessed on 2 September 2026
Jeeva Chacko, ‘Cloning the Persona: Personality and Publicity Rights in the Age of AI Deepfakes and Voice Cloning in India’ ( Indian Journal of Jurisprudence and Reviews, 2 June 2026) https://www.eduresearchjournal.com/index.php/ijjr/article/view/788
San Jose, ‘Artificial Intelligence Voice Scams on the Rise with 1 in 4 Adults Impacted’, (Business Wire, 2 May 2023), https://www.businesswire.com/news/home/20230501005587/en
[1] San Jose, ‘Artificial Intelligence Voice Scams on the Rise with 1 in 4 Adults Impacted’, (Business Wire, 2 May 2023), https://www.businesswire.com/news/home/20230501005587/en
