Author: Mishti Pahuja, a 3rd year law student at DME, GGSIPU
Abstract
Artificial Intelligence (AI) has transformed the way digital content is created, shared, and consumed. One of its most controversial developments is deepfake technology, which enables the creation of highly realistic yet fabricated images, videos, and audio recordings. While deepfakes have legitimate applications in education, entertainment, healthcare, and accessibility, they have also become powerful tools for misinformation, identity theft, cybercrime, political manipulation, financial fraud, and reputational harm. As deepfakes become more advanced, they create legal challenges for legal systems around the world, because courts are increasingly relying on electronic records as evidence. Traditionally, photographs, videos, and audio recordings were viewed as reliable forms of proof. Today, however, advances in AI have made it possible to manipulate such material so convincingly that even trained professionals may struggle to distinguish between genuine evidence and fabricated content.
This article examines how deepfake technology threatens the administration of justice in India by undermining the authenticity and reliability of digital evidence. It analyses the existing legal framework under the Bharatiya Sakshya Adhiniyam, 2023, the Bharatiya Nyaya Sanhita, 2023, the Information Technology Act, 2000, and the Digital Personal Data Protection Act, 2023, while also exploring constitutional concerns relating to privacy, free speech, and due process. The article further compares India’s legal position with developments in jurisdictions such as the European Union, the United States, and China. It argues that although Indian law recognises electronic evidence, it remains inadequately prepared to address the unique challenges created by AI-generated synthetic media. The article concludes by proposing practical reforms, including specialised forensic standards, judicial training, platform accountability, and dedicated legislation to preserve public confidence in the justice delivery system in the age of artificial intelligence.
Keywords: Deepfakes, Artificial Intelligence, Digital Evidence, Bharatiya Sakshya Adhiniyam, Electronic Records, Digital Forensics, Indian Judiciary, Administration of Justice.
Introduction
At the heart of every court proceeding lies a deceptively simple question: what actually happened? Courts have historically answered that question through witness testimony, documents, photographs, and audio-visual recordings. Among these, visual and audio evidence carried particular weight. “Seeing is believing” was not merely a cultural stereotype it shaped how judges evaluated facts for generations.
That assumption is now in serious trouble. Artificial intelligence has made it possible to generate photographs that never existed, clone voices from a few seconds of audio, and produce videos showing people doing or saying things they never did. These fabrications commonly called deepfakes are not the product of clumsy editing. Powered by deep learning and Generative Adversarial Networks (GANs), they are often indistinguishable from real footage, even to trained professionals.
For Indian courts, this is not a distant problem. CCTV footage, WhatsApp messages, mobile recordings, GPS data, and emails now routinely decide the outcome of criminal trials and civil disputes. The Bharatiya Sakshya Adhiniyam, 2023 (BSA) recognises electronic records as admissible evidence but it was never designed to address AI-generated content that satisfies every formal requirement while being entirely fabricated. This article examines how deepfakes threaten the reliability of digital evidence in India, assesses the adequacy of existing law under the BSA, the Bharatiya Nyaya Sanhita, 2023 (BNS, the Information Technology Act, 2000 (IT Act), and the Digital Personal Data Protection Act, 2023 (DPDPA), explores constitutional tensions around privacy and free expression, surveys international responses, and proposes practical reforms.
Understanding Deepfake Technology and Its Impact on Digital Evidence
The word “deepfake” blends “deep learning” with “fake.” A GAN works like a game between two AI systems. One, called the generator, creates fake images, videos, or audio, while the other, the discriminator, acts like a detective trying to spot what is fake. They keep challenging each other in a continuous loop, helping the generator become better at producing highly realistic synthetic media. Over thousands of cycles, the generator learns to fool the discriminator, and the resulting output becomes progressively more convincing. Modern systems require nothing more than a smartphone, a publicly available app, and a handful of photographs.
The technology is not inherently malicious. Film studios use it to de-age actors or restore damaged archival footage. Voice synthesis allows people who have lost the ability to speak to communicate using digitally reconstructed voices. Museums use AI to recreate historical figures and events. These are genuine and valuable applications. The problem is that the same tools are freely accessible to bad actors who can use them to fabricate evidence, harass individuals, and manipulate public opinion.
The documented harms are serious. Non-consensual intimate deepfakes overwhelmingly targeting women cause lasting psychological and reputational damage. AI voice cloning has been used to impersonate executives and defraud companies of millions. Fabricated political speeches circulate faster than fact-checkers can respond, shaping electoral outcomes before corrections reach the same audience.
For the justice system, a specific danger has emerged that scholars call the “liar’s dividend.” As deepfake awareness grows, individuals caught on genuine recordings can simply claim the footage is AI-generated triggering doubt where none would previously have existed. The result is a double threat: fabricated evidence may be believed, and authentic evidence may be disbelieved. Both outcomes undermine the court’s ability to find the truth.
Digital Evidence and the Challenges Before Indian Courts
India’s courts have steadily embraced digital evidence. The BSA treats electronic records as documentary evidence, subject to conditions ensuring their authenticity and integrity. The Supreme Court has reinforced this framework through a series of important decisions. In Anvar P.V. v. P.K. Basheer, (2014) 10 S.C.C. 473. Basheer, the Court held that electronic records must satisfy prescribed legal requirements before admission. In Arjun Panditrao Khotkar v. Kailash Kushanrao Gorantyal, (2020) 7 S.C.C., it reaffirmed that compliance with statutory authentication requirements is generally mandatory. These rulings strengthened procedural safeguards and placed the burden squarely on the party producing electronic evidence to establish its reliability.
But procedural compliance alone cannot now guarantee authenticity. A deepfake video can satisfy every formal evidentiary requirement it can be presented with a valid certificate, intact metadata, and an unbroken chain of custody while still depicting events that never occurred. The legal question has therefore shifted from ‘was this record properly preserved?’ to ‘does this content actually reflect reality?’ The BSA, following the framework of the Indian Evidence Act it replaced, was not designed to answer the latter question.
The practical burden on judges is already significant. Courts must evaluate competing expert opinions on metadata analysis, pixel irregularities, voice frequency patterns, and AI detection software highly technical matters that most legal professionals have not been trained to assess. The shortage of adequately equipped forensic laboratories at the state level compounds the problem. Without uniform national standards for examining synthetic media, the reliability of expert opinions will inevitably vary, introducing another layer of unpredictability into criminal and civil proceedings.
There is also what one might call the arms race problem. Detection tools that reliably identify deepfakes today may be ineffective against more sophisticated AI models released next year. Forensic methods are reactive by nature; they respond to technologies that already exist rather than those being developed. Courts cannot wait for detection science to catch up with generation science. The legal framework must therefore be structured in a way that does not depend entirely on the temporary superiority of any detection methodology.
Existing Legal Framework in India
India does not yet have a dedicated deepfake law, but several existing statutes offer partial remedies. The IT Act, 2000 addresses identity theft, hacking, and unauthorised access, while the IT (Intermediary Guidelines and Digital Media Ethics Code) Rules, 2021 require platforms to exercise due diligence and establish takedown mechanisms for unlawful content. The BNS, 2023 covers cheating by impersonation, criminal defamation, voyeurism, obscenity, and criminal intimidation all of which may apply to deepfake scenarios depending on the circumstances.
Constitutional jurisprudence adds an important layer. The Supreme Court’s landmark judgment in Justice K.S. Puttaswamy (Retd.) v. Union of India, (2017) 10 S.C.C. recognised privacy as a fundamental right under Article 21 and affirmed individual autonomy over personal identity. Deepfakes which reproduce a person’s face, voice, or likeness without consent directly violate this autonomy. R. Rajagopal v. State of Tamil Nadu, (1994) 6 S.C.C. 632 similarly held that individuals have a right to control publication of matters relating to their private lives. Building on this foundation, Indian High Courts have recently extended personality rights protection to public figures including Amitabh Bachchan, Anil Kapoor, Arijit Singh, and Jackie Shroff, recognising that unauthorised digital exploitation of a person’s identity can attract court-ordered relief and damages. The principle underlying those decisions that every person has an interest in controlling their own image and identity applies equally to ordinary victims of malicious deepfakes.
The DPDPA, 2023 further strengthens individual control over personal data by emphasising consent and lawful processing. Since AI systems require enormous datasets of facial images and voice recordings to train deepfake models, data protection principles may indirectly constrain certain aspects of their creation. However, the Act does not address synthetic media directly or prescribe obligations specific to AI-generated impersonation.
Taken together, this framework demonstrates genuine legal creativity existing provisions can be stretched to address many individual deepfake harms. But they were drafted before generative AI became widespread, they require courts to apply concepts like ‘personation’ and ‘publication’ to synthetic media that never captured real events, and they create no specialised standards for evaluating AI-generated evidence in judicial proceedings. The result is a patchwork that addresses consequences rather than causes.
Constitutional Dimensions: Privacy, Free Speech, and Fair Trial
Regulating deepfakes involves balancing important constitutional rights and challenges. The right to privacy under Article 21 is clearly engaged when someone’s face or voice is used without consent to fabricate harmful content. So is the right to dignity, which strengthens the entire framework of fundamental rights. These values point strongly towards regulation.
Article 19(1)(a) protects the right of people to express themselves freely, including through art, satire, political opinions, historical storytelling, and parody. AI can also be used for useful and lawful purposes, like making educational content, documentaries, or movie effects. So, banning all AI-generated or synthetic media completely would be unfair because it would also stop these legitimate uses. Such a broad ban would likely fail the “proportionality test” in the Justice K.S. Puttaswamy v. Union of India case, which says that restrictions on rights must be reasonable and not excessive.
The right to a fair trial also rooted in Article 21 adds a third dimension. Courts are constitutionally obliged to base their decisions on reliable evidence. If fabricated media is admitted without proper authentication, innocent persons risk wrongful conviction. If genuine evidence is rejected because courts distrust all digital recordings, victims are denied justice. Both failures are constitutionally unacceptable. The framework must therefore be proportionate: targeted at harmful and non-consensual uses of synthetic media, rather than AI-generated content as such.
Comparative International Approaches
Several jurisdictions have recognised that existing law contains gaps and have moved towards dedicated regulation. The European Union’s Artificial Intelligence Act classifies AI systems by risk level and requires transparency where synthetic or manipulated content may mislead the public, enabling audiences to distinguish between authentic and fabricated content. It is oriented towards preventing harm before it occurs rather than merely punishing it afterwards.
China has implemented one of the strictest mandatory labelling regimes, requiring synthetic media to carry clear disclosures and placing traceability obligations on service providers. The United States has adopted a decentralised model: multiple states have enacted laws targeting election-related and non-consensual intimate deepfakes, while comprehensive federal legislation remains limited. American legal scholarship has extensively developed the concept of the liar’s dividend, warning that even genuine recordings are becoming easier to challenge as public awareness of deepfake technology grows.
At the international level, the UNESCO Recommendation on the Ethics of Artificial Intelligence (2021) emphasises transparency, accountability, human oversight, and protection of fundamental rights. While not legally binding, these principles have influenced national regulatory conversations worldwide. The common thread across jurisdictions is a growing recognition that traditional legal principles developed for human actors operating on physical materials may not by themselves adequately govern AI-generated synthetic media.
Recommendations
Addressing the deepfake problem in India will require simultaneous action across several fronts. First, Parliament should enact dedicated legislation specifically governing synthetic media. The law should define deepfakes precisely, distinguish clearly between lawful and harmful uses, assign civil and criminal liability to creators and distributors, and impose obligations on AI developers and online platforms. Criminalising fabrication alone will not be enough if platforms continue to strengthen harmful content without consequence.
Second, the BSA should be supplemented with AI-specific evidentiary guidelines. Courts should be empowered and encouraged to require forensic authentication of disputed audio or video recordings before attaching significant weight to them. Uniform national standards for examining synthetic media, developed in consultation with forensic experts and technologists, would bring much-needed consistency to proceedings across different states.
Third, India must invest in forensic infrastructure. State-level laboratories remain inadequately equipped for AI-era investigations. Expanding capacity, standardising protocols, and integrating AI-based detection software, blockchain verification, and watermarking technologies would significantly improve investigative quality and courtroom reliability.
Fourth, judicial and legal training must be treated as urgent rather than optional. Judges, prosecutors, defence lawyers, and police officers all need regular, structured training on AI technologies, digital forensics, and the evidentiary challenges they create. Good laws administered by poorly informed decision-makers will produce poor outcomes.
Finally, public digital literacy deserves serious attention. Citizens who understand what deepfakes are, how to verify digital content, and what legal remedies exist are more resilient to manipulation. No regulatory framework will be fully effective if the population it protects cannot critically evaluate what it sees online.
Conclusion
When I started researching this article, I assumed the problem was primarily a cybercrime issue something addressed by updating the IT Act or adding a few new sections to the BNS. The deeper I looked, the more I realised it is something more fundamental: a challenge to the epistemological foundation of judicial proceedings. Courts have always worked on the assumption that evidence however imperfect attempts to represent reality. Deepfakes sever that relationship entirely.
India has built a solid foundation through the BSA’s treatment of electronic records, the constitutional privacy jurisprudence flowing from Puttaswamy, the personality rights doctrine developed through cases involving Amitabh Bachchan, Anil Kapoor, Arijit Singh, and Jackie Shroff, and the procedural rigour mandated by Anvar P.V. and Arjun Panditrao. These are genuine achievements. But they were not designed for a world where a convincing video of someone committing a crime can be produced in minutes by anyone with a smartphone.
The answer is not to treat AI as an enemy of justice. The same technologies that generate deepfakes also power diagnostic medicine, climate modelling, and criminal investigation. The answer is to ensure that legal institutions evolve alongside technological capabilities through dedicated legislation, stronger forensic infrastructure, informed judges, accountable platforms, and a public that understands what it is looking at.
The administration of justice has always been about distinguishing truth from falsehood under conditions of uncertainty. In the age of artificial intelligence, that task has become harder, but it has not become impossible provided the legal system takes the challenge seriously before the next generation of deepfake technology makes today’s safeguards obsolete.
Refrences:
- Chesney, Robert & Danielle Keats Citron, Deep Fakes: A Looming Challenge for Privacy, Democracy, and National Security, 107 California Law Review 1753 (2019).
- Mirsky, Yisroel & Wenke Lee, The Creation and Detection of Deepfakes: A Survey, 54(1) ACM Computing Surveys, Article 7 (2021).
- Goodfellow, Ian, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair, Aaron Courville & Yoshua Bengio, Generative Adversarial Nets, Advances in Neural Information Processing Systems (NeurIPS 2014).
- UNESCO, Recommendation on the Ethics of Artificial Intelligence (2021).
