Can AI Be Cross-Examined? Examining the Evidentiary Challenges of AI-Generated Evidence in Indian Courts

INTRODUCTION

Imagine a criminal trial in which a CCTV video appears to show an individual accepting a bribe, or a voice recording seemingly captures a confession to a crime. Traditionally, such evidence would be regarded as highly convincing[1]. However, in the age of artificial intelligence, videos, photographs, and voice recordings can be created or manipulated with remarkable accuracy through deepfake technology and generative AI tools[2]. As a result, courts are increasingly confronted with a difficult question: can evidence that may have been generated by software be trusted in the same way as evidence created by humans[3]?

Indian law has gradually adapted to the digital era through the Bharatiya Sakshya Adhiniyam, 2023 and the Information Technology Act, 2000, both of which recognize electronic evidence. Yet these laws were primarily designed for conventional electronic records such as emails, CCTV footage, and digital documents. They provide limited guidance on how courts should evaluate the admissibility, authenticity, and reliability of AI-generated evidence, particularly when the reasoning behind an AI system's output remains opaque[4].

Against this backdrop, this blog examines whether the existing Indian legal framework is capable of determining the admissibility, authenticity, and reliability of AI-generated evidence, including deepfakes, synthetic media, and AI-created digital records, while safeguarding the constitutional guarantee of a fair trial.

EXISTING LEGAL FRAMEWORK IN INDIA

India does not yet have a specific law that lays down how to approach the subject of AI-generated evidence in India yet. As such, the courts are still dealing with evidence as it stands now, since they are bound to follow the existing framework governing electronic and digital evidence, primarily under the Bharatiya Sakshya Adhiniyam, 2023 (BSA). The BSA is the legislation that acknowledges electronic records as admissible evidence and defines the basis of admission in court for digital materials such as emails, CCTV footage, e-Documents, audio recordings and computer-generated records.[5]

Several provisions of the BSA become relevant when AI-generated evidence is produced before a court. The provisions relating to electronic records and electronic evidence require parties to establish the authenticity and integrity of digital material.[6] The law also recognizes expert opinion as relevant evidence, which means that forensic experts may be called upon to verify the authenticity of AI-generated content such as deepfake videos, cloned voices, or algorithmic reports. In addition, the provisions governing examination and cross-examination of witnesses remain important because parties have a right to challenge the evidence presented against them.[7]

The Information Technology Act, 2000, also plays an important role. It grants legal recognition to electronic records and electronic signatures and forms the backbone of India's digital evidence regime.[8] Together, the IT Act and the BSA create the legal framework through which courts assess electronic evidence.

The Supreme Court's decisions in Anvar P.V. v. P.K. Basheer and Arjun Panditrao Khotkar v. Kailash Kushanrao Gorantyal clarified that electronic evidence must satisfy statutory certification requirements before it can be admitted.[9]Similarly, Tomaso Bruno v. State of Uttar Pradesh highlighted the growing importance of electronic evidence in criminal investigations.[10]

However, these laws and court decisions were developed for traditional electronic evidence and do not specifically cover AI-generated content. As a result, while we can currently evaluate AI-generated material using existing rules on electronic evidence, the ability of these provisions to effectively handle new technologies like deepfakes, synthetic media, and AI-assisted forensic tools is still an ongoing legal issue.[11]

ANALYSIS

Admissibility: Can AI Evidence Even Enter the Courtroom?

Under the Bharatiya Sakshya Adhiniyam, 2023, AI-generated evidence would generally be treated as electronic evidence. The provisions governing electronic records, together with the principles laid down in Anvar P.V. v. P.K. Basheer and Arjun Panditrao Khotkar v. Kailash Kushanrao Gorantyal, require proof of authenticity and statutory certification before such evidence can be admitted.[12]

However, these provisions were designed for traditional electronic records such as emails, CCTV footage, call records, and digital documents. They do not specifically address AI-generated content.[13] For example, if a deepfake video showing a person accepting a bribe is produced before a court, the existing law can verify whether the file was properly obtained and stored, but it does not provide a mechanism to determine whether the video itself was artificially generated. As a result, the law adequately addresses admissibility but fails to fully address the unique nature of AI-created evidence.

Authenticity: Is the Evidence Genuine?

Authenticity is one of the most significant challenges posed by AI-generated evidence. Traditional digital evidence can often be verified through metadata, device records, or chain-of-custody documentation. However, modern AI tools can create highly realistic videos, photographs, and voice recordings that appear completely genuine.

The recent occurrence of voices of political leaders that have been cloned through AI technology and deep fake videos of public personalities has proven that it is possible to create something realistic using artificial intelligence. In case the audio or video was created as evidence during a criminal court proceeding, it would be difficult for the court to ascertain if the evidence was genuine or not."[14]Even if the file is technically authentic as a digital record, the content itself may be fabricated. This demonstrates a gap in the current legal framework. Neither the Bharatiya Sakshya Adhiniyam nor the Information Technology Act, 2000, contains specific provisions addressing deepfakes or synthetic media.[15] The result is an "authenticity crisis" in which courts may struggle to distinguish between genuine and manipulated evidence[16].

Reliability: Can Courts Trust AI Outputs?

Even if AI-generated evidence is admissible and appears authentic, courts must still determine whether it is reliable. Reliability refers to the accuracy and trustworthiness of the process that produced the evidence.[17] This issue arises particularly in facial-recognition systems, predictive policing algorithms, and AI-assisted forensic tools. Research has shown that certain facial-recognition technologies may produce higher error rates for women, elderly individuals, and persons with darker skin tones.[18]

Suppose a facial-recognition system identifies a suspect from CCTV footage with 90% confidence. Should a court rely on that output? Can the defence examine the training data used by the AI system? Can the software developer explain how the conclusion was reached?

The current Indian evidence law contains provisions with respect to expert opinions, but does not provide a specific test for determining the reliability of AI systems. The US court employs the Daubert standard to determine the reliability of the scientific and technological evidence. There are no provisions in India with respect to testing the reliability of AI-based evidence. According to the Daubert standard, the judge considers criteria such as whether the method is testable, whether it is peer-reviewed, the error rate, and whether the scientific community has accepted it. Given the issues related to inaccuracies in the data used to train the algorithms and other issues, it becomes imperative to have some sort of test for reliability in this case. In the absence of this test, it will not be possible for Indian courts to decide on the reliability of the AI technology generating the evidence.[19]

Fair Trial and Constitutional Concerns

The most serious issue is whether reliance on AI-generated evidence is compatible with the constitutional guarantee of a fair trial under Article 21 of the Constitution.[20] The Indian criminal justice system is based on the adversarial model, where evidence is tested through cross-examination. However, AI systems cannot be cross-examined in the traditional sense. A machine cannot explain its reasoning, take an oath, or answer questions regarding potential bias. This becomes particularly problematic when courts rely on "black-box" algorithms whose internal functioning is not publicly disclosed. If an accused person is convicted on the basis of a deepfake analysis tool, facial-recognition software, or predictive risk assessment system, the defence may be unable to meaningfully challenge the evidence. This undermines principles of natural justice and procedural fairness.[21] The concern is not merely technological; it is constitutional. If parties cannot effectively challenge AI-generated evidence, the right to a fair trial may be compromised.[22]

Comparative Perspective

Compared to India, some other jurisdictions like the USA, UK and European Union have started to formulate certain measures to counter the threats posed by artificial intelligence-generated evidence[23]. In the United States, courts use the Daubert standard to evaluate the reliability of scientific and technological evidence before they allow it in court.[24] Similarly, the European Union's AI Act requires transparency, documentation, and human oversight for high-risk AI systems[25], including those in law enforcement and judicial administration.[26]

In contrast, India's legal framework under the Bharatiya Sakshya Adhiniyam, 2023, mainly focuses on the authenticity of electronic records instead of the reliability of the AI systems that create them. While Indian courts can check if a digital record has been properly certified, there is no clear way to assess algorithmic bias, error rates, explainability, or the accuracy of AI-generated outputs. As a result, India could benefit from clearer standards on AI reliability, transparency, and independent forensic verification, all while ensuring that technological advancements uphold the constitutional guarantee of a fair trial.[27]

THE WAY FORWARD

The rise of deepfake videos, AI-generated voice recordings, facial-recognition systems, and automated forensic tools shows that current evidence laws are not equipped to handle the challenges from artificial intelligence. Although the Bharatiya Sakshya Adhiniyam, 2023, offers a basis for electronic evidence, it fails to fully address issues of transparency, authenticity, reliability, and accountability.

Moving forward, courts may need to implement stricter verification processes before accepting AI-generated evidence. For instance, deepfake videos and cloned audio recordings should be examined by independent forensic experts. Additionally, AI systems used by law enforcement must reveal their accuracy rates and any potential biases.[28] Judges and judicial officers should also receive technical training to better grasp these new technologies. The Indian legal system should embrace AI instead of shying away from it. These actions would not stop AI from being used in the legal system; rather, they would help ensure that technological advances support, not weaken, the values of fairness, accuracy, and due process that underpin Indian evidence law.[29]

CONCLUSION

The rapid rise of artificial intelligence has fundamentally challenged traditional evidence law principles. The existing Indian legal framework to admit electronic evidence under the Bharatiya Sakshya Adhiniyam, 2023, and the Information Technology Act, 2000[30]. It, however, cannot help the nation address the emerging issues associated with AI-generated outputs such as deepfakes, synthetic media, voice cloning, facial recognition, and automated forensic reports. Although the law can often determine if a digital record is authentic and properly certified, it is less equipped to evaluate how AI-generated outputs are made, whether they are unbiased, and if they can be effectively challenged in court.

From the above discussion, it becomes evident that the problem of admitting AI-generated evidence in Indian courts lies not in technology but in constitutionality. The increasing use of AI-generated outputs as evidence is raising serious questions regarding the transparency, accountability, and the right to a fair trial under Article 21 of the Constitution of India. Thus, the key question here is not whether AI-generated evidence should be admitted in Indian courts, but whether there are adequate legal mechanisms in place to ensure the justice and legality of such evidence. As the artificial intelligence continues to evolve, Indian evidence law also needs to evolve.

REFERNCES

[1] Tomaso Bruno v. State of Uttar Pradesh, (2015) 7 S.C.C. 178.

[2] Danielle Keats Citron & Robert Chesney, Deep Fakes: A Looming Challenge for Privacy, Democracy, and     National Security, 107 Calif. L. Rev. 1753 (2019).

[3] Ananya Sharma & Ekta Gupta, AI-generated Evidence in the Indian Legal System: Navigating the Evidentiary Vacuum and Charting a Reform Agenda, 6 Indian Journal of Integrated Research in Law, Issue II (2025)

[4] Sharma & Gupta, supra note 7, at 2973–2975.

[5] Bharatiya Sakshya Adhiniyam, No. 47 of 2023, §§ 57–63, India Code (2023).

[6] Bharatiya Sakshya Adhiniyam, No. 47 of 2023, §§ 61–63, India Code (2023).

[7] Bharatiya Sakshya Adhiniyam, No. 47 of 2023, §§ 39–45, India Code (2023).

[8] Information Technology Act, No. 21 of 2000, §§ 4–5, India Code (2000).

[9] Anvar P.V. v. P.K. Basheer, (2014) 10 S.C.C. 473; Arjun Panditrao Khotkar v. KAIlash Kushanrao Gorantyal, (2020) 7 S.C.C. 1.

[10] Tomaso Bruno v. State of Uttar Pradesh, (2015) 7 S.C.C. 178.

[11] Sharma & Gupta, supra note 3, at 2969, 2977–2978.

[12] Bharatiya Sakshya Adhiniyam, No. 47 of 2023, §§ 61–63, India Code (2023); Anvar P.V. v. P.K. Basheer, (2014) 10 S.C.C. 473; Arjun Panditrao Khotkar v. KAIlash Kushanrao Gorantyal, (2020) 7 S.C.C. 1.

[13] Ananya Sharma & Ekta Gupta, AI-generated Evidence in the Indian Legal System: Navigating the Evidentiary Vacuum and Charting a Reform Agenda, 6 Indian J. Integrated Rsch. L., Issue II, 2974–2975 (2025).

[14] Danielle Keats Citron & Robert Chesney, Deep Fakes: A Looming Challenge for Privacy, Democracy, and National Security, 107 Calif. L. Rev. 1753, 1758–1765 (2019); Sharma & Gupta, supra note 3, at 2972–2978.

[15] Information Technology Act, No. 21 of 2000, India Code (2000); Bharatiya Sakshya Adhiniyam, No. 47 of 2023, India Code (2023).

[16] Danielle Keats Citron & Robert Chesney, Deep Fakes: A Looming Challenge for Privacy, Democracy, and National Security, 107 Calif. L. Rev. 1753, 1770–1775 (2019); Sharma & Gupta, supra note 3, at 2977–2978.

[17] Sahibpreet Singh & Lalita Devi, Reliability and Admissibility of AI-generated Forensic Evidence in Criminal Trials 45–49 (2025).

[18] Joy Buolamwini & Timnit Gebru, Gender Shades: Intersectional Accuracy Disparities in Commercial Gender Classification, Proc. FAT* Conf. 2018

[19] Daubert v. Merrell Dow Pharm., Inc., 509 U.S. 579, 593–95 (1993); Ananya Sharma & Ekta Gupta, supra note 3, at 2978–2979.

[20] INDIA CONST. art. 21; K.S. Puttaswamy v. Union of India, (2017) 10 S.C.C. 1.

[21] Ananya Sharma & Ekta Gupta, supra note 2, at 2976–2977; State v. Loomis, 881 N.W.2d 749 (Wis. 2016).

[22] INDIA CONST. art. 21; Maneka Gandhi v. Union of India, (1978) 1 S.C.C. 248.

[23] Daubert v. Merrell Dow Pharm., Inc., 509 U.S. 579 (1993); Regulation (EU) 2024/1689 (Artificial Intelligence Act); Judiciary of England and Wales, Guidance for Judicial Office Holders on the Responsible Use of Artificial Intelligence (2023).

[24] Daubert v. Merrell Dow Pharm., Inc., 509 U.S. 579 (1993)

[25] State v. Loomis, 881 N.W.2d 749 (Wis. 2016)

[26] Regulation (EU) 2024/1689 of the European Parliament and of the Council of 13 June 2024 (Artificial Intelligence Act)

[27] Ananya Sharma & Ekta Gupta, supra note 19, at 2980–2982; INDIA CONST. art. 21

[28] European Parliament and Council Regulation (EU) 2024/1689 (Artificial Intelligence Act); Ananya Sharma & Ekta Gupta, supra note 16, at 2979–2980.

[29] INDIA CONST. art. 21; Maneka Gandhi v. Union of India, (1978) 1 S.C.C. 248; K.S. Puttaswamy v. Union of India, (2017) 10 S.C.C. 1.

[30] Bharatiya Sakshya Adhiniyam, No. 47 of 2023, India Code (2023); Information Technology Act, No. 21 of 2000, India Code (2000).

[1] Danielle Keats Citron & Robert Chesney, Deep Fakes: A Looming Challenge for Privacy, Democracy, and National Security, 107 Calif. L. Rev. 1753, 1758–1765 (2019); Sharma & Gupta, supra note 3, at 2972–2978.

[2] Information Technology Act, No. 21 of 2000, India Code (2000); Bharatiya Sakshya Adhiniyam, No. 47 of 2023, India Code (2023).

[3] Danielle Keats Citron & Robert Chesney, Deep Fakes: A Looming Challenge for Privacy, Democracy, and National Security, 107 Calif. L. Rev. 1753, 1770–1775 (2019); Sharma & Gupta, supra note 3, at 2977–2978.

[4] Sahibpreet Singh & Lalita Devi, Reliability and Admissibility of AI-generated Forensic Evidence in Criminal Trials 45–49 (2025).

[1] Ananya Sharma & Ekta Gupta, AI-generated Evidence in the Indian Legal System: Navigating the Evidentiary Vacuum and Charting a Reform Agenda, 6 Indian J. Integrated Rsch. L., Issue II, 2974–2975 (2025).

[1] Bharatiya Sakshya Adhiniyam, No. 47 of 2023, §§ 61–63, India Code (2023); Anvar P.V. v. P.K. Basheer, (2014) 10 S.C.C. 473; Arjun Panditrao Khotkar v. KAIlash Kushanrao Gorantyal, (2020) 7 S.C.C. 1.

[1] Information Technology Act, No. 21 of 2000, §§ 4–5, India Code (2000).

[2] Anvar P.V. v. P.K. Basheer, (2014) 10 S.C.C. 473; Arjun Panditrao Khotkar v. KAIlash Kushanrao Gorantyal, (2020) 7 S.C.C. 1.

[3] Tomaso Bruno v. State of Uttar Pradesh, (2015) 7 S.C.C. 178.

[4] Sharma & Gupta, supra note 3, at 2969, 2977–2978.

[1] Bharatiya Sakshya Adhiniyam, No. 47 of 2023, §§ 57–63, India Code (2023).

[2] Bharatiya Sakshya Adhiniyam, No. 47 of 2023, §§ 61–63, India Code (2023).

[3] Bharatiya Sakshya Adhiniyam, No. 47 of 2023, §§ 39–45, India Code (2023).

[1] Danielle Keats Citron & Robert Chesney, Deep Fakes: A Looming Challenge for Privacy, Democracy, and     National Security, 107 Calif. L. Rev. 1753 (2019).

[2] Ananya Sharma & Ekta Gupta, AI-generated Evidence in the Indian Legal System: Navigating the Evidentiary Vacuum and Charting a Reform Agenda, 6 Indian Journal of Integrated Research in Law, Issue II (2025)

[3] Sharma & Gupta, supra note 7, at 2973–2975.

[1] Tomaso Bruno v. State of Uttar Pradesh, (2015) 7 S.C.C. 178.