Author: Mayank Chaudhary, Fairfield Institute of Management and Technology
Abstract
Courts have long treated recordings as strong proof. Generative software now makes it cheap to produce audio and video of events that never occurred, which puts pressure on that habit. This article describes the resulting dilemma, maps the Indian laws that already touch it, identifies the point at which the current evidence rules stop helping, and offers two original suggestions: a three-step method for examining suspect media and a two-stage procedure for handling claims of fabrication. It closes with brief international comparisons and a set of practical questions and answers.
Keywords: deepfakes, synthetic media, electronic evidence, authenticity, expert opinion, personality rights
1. A custody hearing, imagined
In a family court, a mother seeks custody of her ten-year-old son. Her lawyer plays a sixty-second audio clip in which the father seems to threaten the child. The father insists he never said it and that someone has copied his voice from videos he posted online.
The clip arrives with a neat certificate stating that it was transferred from the mother’s phone without change. That may well be true. Yet the certificate cannot answer the only question that matters: was the voice ever the father’s?
The scenario is invented, but it shows why the issue is urgent. A family, a business or a defendant may rise or fall on a recording, and the old confidence in recordings can no longer be assumed.
Statements about law and events below reflect the author’s understanding and should be checked against the sources at the end.
2. Doubt cuts both ways
2.1 Believing a forgery
One risk is straightforward. A convincing fake enters the record and tips a decision. Voice-cloning tools need only a short sample, and image tools can graft one person’s face onto another’s body. Misuse outside courts has been reported, such as a 2024 Hong Kong incident in which an employee was said to have been misled by a fabricated video meeting into sending a large payment.
2.2 Doubting the truth
The other risk is less obvious. Scholars Robert Chesney and Danielle Citron call it the “liar’s dividend”: because fakes exist, a person caught by genuine evidence can say it was manufactured. A reported 2023 US civil case involving a technology executive is often cited, in which an argument that statements might be deepfakes was said to have been rejected by the judge as unsupported.
A court that is easily deceived harms the innocent. A court that can be stalled by a casual claim of fakery protects the guilty. Any workable approach must guard against both.
3. What Indian law already offers
Most of these tools treat deepfakes as wrongs to be punished or removed. They say little about the narrower question a judge faces mid-trial: what to do with a disputed clip that has already been placed on the record.
4. The point where the rules stop
4.1 A certificate describes a journey, not a birth
Section 63 certifies that a file on a device is the file that was produced from it. A synthetic clip that was created elsewhere and then copied to a phone can carry a flawless certificate, because the certificate begins its story after the clip already existed.
4.2 Older rules on tapes still point the way
The approach to tape recordings associated with Ram Singh v. Col. Ram Singh (1985) asks courts to check that the speaker is identified, that the recording is accurate and untampered, and that its content is relevant. These ideas remain useful as a starting point, although they were framed for analogue media and not for software that generates voices.
4.3 Experts help, but they are not oracles
Forensic examiners can look for odd lighting, unnatural eye movement, audio artefacts and compression traces. But detection and generation are locked in a race, and a test that works today may fail next year. Experts can also disagree, and a judge without technical training may struggle to choose between them.
5. A three-step method: trace, test, triangulate
The following method is the author’s own suggestion, not a legal test.
1. Trace. Establish who first created or received the file, on which device and account, and whether the original source can be produced. Metadata, upload histories and content-credential information, where it exists, can help rebuild the path.
2. Test. Ask an independent expert to analyse the file for manipulation and to compare voices or faces with known genuine samples. The expert should explain the method, the error rate and what could not be determined.
3. Triangulate. Compare the clip with everything else: witness accounts, phone location records, access logs, messages and the surrounding circumstances. A clip that fits independent facts gains credibility, while one that clashes with reliable records loses it.
In the custody example, the court would ask who recorded the clip and when, whether the original phone shows the recording, what a voice expert concludes, and whether the child’s own statements, messages or other witnesses support the claimed threat.
6. A two-stage way to handle fake claims
To avoid both gullibility and easy escape, the author suggests that courts consider the following procedure. It is a reform proposal and not existing law.
Stage one: give reasons. A party alleging a deepfake should explain why, for instance by pointing to inconsistencies, a motive to fabricate or technical oddities, instead of simply asserting it.
Stage two: show the source. If the challenge is reasoned, the party relying on the clip must demonstrate its origin and handling, and the court may order expert examination. Where origin cannot be shown and tests are inconclusive, the court may give the clip less weight or require other proof before relying on it.
The ordinary burden of proof stays where it is. The point is that unsupported suspicion and unsupported confidence both receive a test.
7. Glimpses from abroad
In the United States, rule-makers have been considering whether the federal evidence rules need updating for machine-generated and AI-altered material, and the topic is under active discussion. The European Union’s Artificial Intelligence Act contains transparency duties that require disclosure when content has been artificially generated or manipulated. China adopted rules in 2023 requiring labelling of synthetic content from deep synthesis services.
These measures concentrate on prevention and disclosure. They do not replace courtroom methods for deciding what to do with a contested recording. Reports indicate that India has also considered labelling requirements for synthetic media, and readers should confirm the current position.
8. Taking stock
8.1 Strengths of the current position
• Judicial flexibility. Courts can already call experts, ask for original devices and approach evidence with caution.
• Fast remedies for victims. Takedown directions and personality-rights injunctions can act quickly.
• A modernised code. The BSA expressly recognises electronic records and standardises the certificate.
8.2 Weak spots
• Capacity. Forensic labs and trained examiners are scarce, especially away from major cities.
• Time and cost. Close examination of each disputed clip burdens parties and slows trials.
• A moving target. Detection methods age quickly.
• No settled threshold. Indian law has not yet fixed how much proof of origin is required once fabrication is alleged.
9. Conclusion
The custody hearing shows the challenge in miniature. The court need not choose between trusting the clip and rejecting it. It can ask where the clip came from, what independent analysis shows and whether other evidence points the same way.
India already has pieces of the answer: a certificate regime, expert evidence and strong remedies against misuse. What is missing is an agreed method for the moment authenticity is truly in doubt. Trace, test, triangulate, together with a two-stage procedure for fakery claims, would give courts a calm way to respond: protected against forgery, and not hostage to every excuse.
10. Frequently Asked Questions
Q1. What exactly is a deepfake?
It is audio, video or imagery generated or altered by artificial intelligence so that it appears to show a person saying or doing something they did not.
Q2. Will a court automatically refuse a deepfake?
There is no automatic rule. Courts judge every electronic record for authenticity and relevance, and proven fabrication will not be relied on and may expose the creator to prosecution.
Q3. Does a Section 63 certificate rule out a deepfake?
No. It supports the claim that a file is unchanged since it was captured or copied, but it does not show that the events in it really happened.
Q4. Can a person just say real evidence is fake?
They can say so, but a bare claim should carry little weight. Under the proposal in this article, they would first have to give reasons, and then the other side would have to show the clip’s origin.
Q5. How do experts spot synthetic media?
By examining the file, its metadata and visible or audible irregularities, and by comparing it with genuine samples. Their conclusions are probabilities and should state limitations.
Q6. Which laws may apply to deepfake misuse?
Depending on the facts, IT Act provisions on identity theft, personation, privacy and obscene content, and Bharatiya Nyaya Sanhita provisions on cheating, forgery and defamation. The exact sections should be verified for each case.
Q7. What can a victim do immediately?
Report the content to the platform, file a complaint through cybercrime reporting channels and, where appropriate, seek urgent court orders such as personality-rights injunctions.
Q8. Should courts stop trusting video and audio?
No. These remain valuable. The sensible response is closer scrutiny of origin and corroboration, not blanket suspicion.
References
Cases
• Ram Singh v. Col. Ram Singh, 1985 Supp SCC 611.
• Anvar P.V. v. P.K. Basheer, (2014) 10 SCC 473.
• Arjun Panditrao Khotkar v. Kailash Kushanrao Gorantyal, (2020) 7 SCC 1.
• Anil Kapoor v. Simply Life India, Delhi High Court (2023).
• Arijit Singh v. Codible Ventures LLP, Bombay High Court (2024).
Statutes and instruments
• Bharatiya Sakshya Adhiniyam, 2023, sections 39 and 63 and the Schedule.
• Information Technology Act, 2000, sections 66C, 66D, 66E, 67 and 67A.
• Bharatiya Nyaya Sanhita, 2023 (provisions on forgery, personation and defamation).
• Information Technology (Intermediary Guidelines and Digital Media Ethics Code) Rules, 2021, and related government advisories.
• Regulation (EU) 2024/1689 (Artificial Intelligence Act), transparency provisions.
Scholarship
• R. Chesney and D. Citron, “Deep Fakes: A Looming Challenge for Privacy, Democracy, and National Security”, 107 California Law Review 1753 (2019).
Note: Case details, section numbers and descriptions of recent developments and incidents are given to the best of the author’s knowledge and should be checked against official sources and current reporting before use in academic or professional work.
This article is for academic and educational purposes and is not legal advice.
