Artificial Intelligence & the Future of Evidence Adjudication: Recalibrating The Bharatiya Sakshya Adhiniyam, 2023 for the Algorithmic Age

This is an Award-winning Essay conferred with a Certificate of Rank-based Merit by LegalWiki as part of LegalWiki's 1st National Article Writing Competition.

---

Artificial Intelligence is no longer a peripheral technology in the courtroom; it is fast becoming both a tool of adjudication and a source of the very evidence that courts are called upon to evaluate. This paper examines how the Bharatiya Sakshya Adhiniyam, 2023 (“BSA”), India’s new evidentiary code, is equipped, or ill-equipped, to respond to AI-generated content, algorithmic outputs, and machine-assisted fact-finding. It argues that while the BSA’s expanded definition of “document” and its treatment of electronic and digital records represents a forward-looking attempt to future-proof Indian evidence law, the absence of AI-specific authentication standards leaves adjudicators without a coherent framework to test the provenance, integrity, and reliability of machine-generated evidence. That gap is no longer hypothetical: in January 2025, the Income Tax Appellate Tribunal at Bengaluru recalled an order that rested on four non-existent, AI-generated citations, and in January 2026 the Andhra Pradesh High Court dealt with a trial court order built on AI-fabricated precedents. Drawing on the certificate requirement governing electronic records and the jurisprudence on authentication of electronic evidence developed by the Supreme Court of India, the paper explores the interpretive strain that AI-generated content and deepfakes place on the traditional standards of authenticity and best evidence. It further considers the implications of AI-assisted judicial decision-making for principles of natural justice, transparency, and reasoned adjudication. The research article concludes with concrete recommendations: statutory recognition of AI-origin evidence as a distinct sub-category, mandatory metadata and provenance certification, judicial training in digital forensics, and a calibrated evidentiary presumption against unauthenticated synthetic media. The paper ultimately contends that the future credibility of adjudication in India depends on evidence law evolving as quickly as the technology it is meant to regulate.

Keywords: Artificial Intelligence, Bharatiya Sakshya Adhiniyam 2023, Electronic Evidence, Deepfakes, Authentication, Algorithmic Adjudication.

Introduction

Law has always lagged technology by a generation, but the gap has rarely been as consequential as it is with Artificial Intelligence (“AI”). Courts today face a dual reality: AI systems generate content that must be assessed as evidence, and AI systems are themselves being deployed as tools of judicial and quasi-judicial decision-making. India’s replacement of the Indian Evidence Act, 1872 with the Bharatiya Sakshya Adhiniyam, 2023[1] offered a rare legislative opportunity to build an evidentiary code fit for this reality. This paper asks whether that opportunity has been meaningfully seized, and what remains to be done as AI-generated evidence, including deepfakes, synthetic audio, and algorithmically produced records, moves from theoretical concern to courtroom occurrence. When a deepfake video of actor Rashmika Mandanna went viral in November 2023, the Ministry of Electronics and Information Technology issued a formal advisory to intermediaries on deepfake content within weeks.[2]

The Statutory Architecture: BSA 2023 and Electronic Evidence

The BSA retains and expands the definition of “document” to expressly include electronic and digital records, thereby carrying forward the trajectory begun by the Information Technology Act, 2000.[3]

Electronic records are treated as primary evidence where produced from proper custody, and the certificate requirement for secondary electronic evidence, the successor to the erstwhile section 65B of the Indian Evidence Act, 1872, is preserved in substantially similar form.[4]

The Supreme Court’s insistence, in Anvar P V v P K Basheer, that a certificate is a mandatory precondition to the admissibility of electronic evidence, later reaffirmed and clarified by a larger bench in Arjun Panditrao Khotkar v Kailash Kushanrao Gorantyal, continues to govern the interpretation of this provision under the new code.[5]

What this architecture does not do, however, is distinguish between electronic records that are mere digital copies of real-world events, such as a photograph or a recorded conversation, and electronic records that are wholly or partly generated, altered, or synthesised by an algorithm. The certificate mechanism was designed to verify the integrity of a copying or transmission process; it was not designed to interrogate whether the events a record depicts ever happened. This is the structural gap AI now exposes.

The premise that a mechanically produced rendition of a real event is presumptively reliable, once properly proved, runs deep in Indian evidence jurisprudence: the Supreme Court accepted a tape-recorded conversation as substantive evidence in R M Malkani v State of Maharashtra[6] long before the BSA or its predecessor addressed electronic records in terms. It is precisely this founding premise, that the record reflects something that actually happened, that AI-generated content now unsettles.

AI as a Challenge to Authenticity: Deepfakes and Synthetic Media

Deepfake technology, using generative adversarial networks and diffusion models[7] to fabricate video, audio, and images increasingly indistinguishable from genuine recordings, has already produced real litigation problems: parties disputing the authenticity of viral clips, allegedly fabricated call recordings used in matrimonial and criminal proceedings, and synthetic voice clones used in fraud. The BSA’s authentication regime, built around custody, hash-value integrity, and certification of the transmission chain, verifies that a file has not been altered in transit; it says nothing about whether the file’s content was authentic at the moment of its creation.

The Court’s reasoning in Tomaso Bruno v State of Uttar Pradesh, which treated withheld or unproduced electronic evidence as capable of triggering adverse inference, illustrates how heavily adjudication has come to depend on the assumed reliability of digital records, an assumption deepfakes directly threaten.[8]

Two evidentiary risks follow. First, the “liar’s dividend”: genuine evidence may be dismissed as fabricated once deepfakes are known to exist, allowing wrongdoers to escape accountability by simply alleging manipulation.[9] Second, courts may lack the forensic capacity, whether in the form of trained examiners, standardised detection tools, or statutory presumptions, to distinguish authentic from synthetic content at the pace litigation demands. Neither risk is adequately addressed by a certification regime built for an era when copying, not creation, was the primary threat to evidentiary integrity.

AI as an Adjudicatory Tool: Algorithmic Decision-Making and Due Process

The second dimension of this inquiry concerns AI’s use by decision-makers themselves, from risk-assessment tools in bail determinations — of the kind the Wisconsin Supreme Court upheld, subject to written warnings about the tool’s proprietary and opaque scoring, in State v Loomis[10] — to AI-assisted drafting of judgments and orders. Even where such use is confined to research or drafting assistance, it raises questions under the principles of natural justice: a litigant is entitled to know the reasoning that produced a decision affecting them, and an opaque algorithmic recommendation, however statistically sound, sits uneasily with the requirement of a reasoned order.[11] The BSA, being an evidentiary rather than a procedural code, does not directly regulate this use, leaving a regulatory vacuum that procedural law and judicial conduct guidelines have yet to fill comprehensively.

This is no longer a hypothetical concern. Indian courts and tribunals have already encountered orders and pleadings resting on case citations fabricated by generative AI tools. The Income Tax Appellate Tribunal at Bengaluru recalled its own order in Buckeye Trust v Principal Commissioner of Income Tax after it emerged that the order relied on four non-existent judicial precedents placed before it without verification.[12] The Andhra Pradesh High Court, in Gummadi Usha Rani v Sure Mallikarjuna Rao, found that a trial court had relied on non-existent, AI-generated citations, which the judicial officer admitted to sourcing from an AI tool in good faith; while the High Court held that the fake citations did not vitiate the order, since its legal reasoning was sound, the Supreme Court, in further proceedings, observed that reliance on AI-fabricated judgments may amount to professional misconduct rather than a mere correctable error.[13] In response, the Kerala High Court became the first in India to issue a formal policy barring judicial officers from using generative AI tools to arrive at findings or draft orders.[14] These episodes confirm that the risk identified in the deepfake context examined above, unauthenticated synthetic content entering the evidentiary or adjudicatory record, is not confined to deepfake video or audio; it extends to the text of judicial reasoning itself, reinforcing the case for the statutory safeguards proposed in the recommendations that follow.

This vacuum is not unique to India. Regulatory efforts elsewhere, principally the transparency and labelling obligations imposed on synthetic and deepfake content under the European Union’s Artificial Intelligence Act,[15] and India’s own policy discourse around a possible Digital India Act[16] and NITI Aayog’s responsible AI framework,[17] signal an awareness of the problem, but none of these instruments currently possess the binding, evidentiary specificity that the BSA could provide. Until such specificity emerges, litigants and adjudicators alike are left to infer standards of disclosure and reasoned explanation from general administrative law principles, an uneasy substitute for a dedicated evidentiary rule addressing algorithmic input into judicial reasoning.

Recommendations for Reform

Four reforms would close that gap. First, the BSA should recognise “AI-generated” or “synthetic” content as a distinct sub-category of electronic record, triggering an enhanced authentication standard rather than the ordinary certificate procedure. Second, provenance and metadata certification, recording the tools, models, or platforms used to create or process a record, should be made a precondition for admissibility of any digital record where AI involvement is alleged or suspected. Third, judicial and forensic infrastructure must be strengthened through mandatory training in digital forensics and the establishment of accredited deepfake-detection laboratories that courts can rely upon, akin to the role already assigned to forensic experts under the Bharatiya Nagarik Suraksha Sanhita, 2023 for offences carrying serious punishment.[18] Fourth, a rebuttable presumption of unreliability could be introduced for unauthenticated synthetic media, shifting the burden onto the party seeking to rely on such content to establish its authenticity, rather than leaving that burden on the party disputing it, consistent with the general architecture of rebuttable presumptions already recognised under the BSA.[19]

Conclusion

The Bharatiya Sakshya Adhiniyam, 2023 took an important step by formally absorbing electronic and digital records into the mainstream of Indian evidence law. But absorption is not the same as adaptation. AI does not merely add another format of evidence to be certified; it destabilises the very assumption, that a recording reflects a real event, on which the certification model was built. Unless the BSA, through subordinate legislation, judicial interpretation, or future amendment, develops standards specific to AI-generated and AI-assisted evidence, Indian courts will continue to apply a nineteenth-century evidentiary logic, refined for the copy-paste era, to a generative one. The future of adjudication in India will depend on closing that gap before it is exploited rather than after, and on legislators treating that task with the same urgency that the underlying technology itself demands.

References

1. Bharatiya Sakshya Adhiniyam 2023 (Act 47 of 2023), assented to on 25 December 2023 and brought into force on 1 July 2024, repealing the Indian Evidence Act 1872. https://www.indiacode.nic.in/handle/123456789/20063

2. Ministry of Electronics and Information Technology, Government of India, advisory to intermediaries on due diligence in relation to deepfake content (26 December 2023), issued weeks after the viral deepfake video of actor Rashmika Mandanna. https://www.pib.gov.in/PressReleaseIframePage.aspx?PRID=1990542; see also BBC News, ‘Rashmika Mandanna calls for action against “scary” deepfake video’ (November 2023) https://www.bbc.com/news/world-asia-india-67305557

3. Bharatiya Sakshya Adhiniyam 2023, s 2(1)(d). https://www.indiacode.nic.in/handle/123456789/20063

4. Bharatiya Sakshya Adhiniyam 2023, s 63; Information Technology Act 2000, s 65B (as it stood under the Indian Evidence Act 1872 regime). https://indiankanoon.org/doc/125020475/

5. Anvar P V v P K Basheer (2014) 10 SCC 473, https://indiankanoon.org/doc/187283766/; Arjun Panditrao Khotkar v Kailash Kushanrao Gorantyal (2020) 7 SCC 1, https://indiankanoon.org/doc/29077914/

6. R M Malkani v State of Maharashtra AIR 1973 SC 157, holding a mechanically recorded conversation admissible where relevant, accurately reproduced, and free of tampering. https://indiankanoon.org/doc/1179783/

7. Ian Goodfellow and others, ‘Generative Adversarial Nets’ in Advances in Neural Information Processing Systems 27 (NIPS 2014), the foundational computer-science account of the generative-adversarial architecture underlying much deepfake technology. https://papers.nips.cc/paper/5423-generative-adversarial-nets

8. Tomaso Bruno v State of Uttar Pradesh (2015) 7 SCC 178. https://indiankanoon.org/doc/193239104/

9. Robert Chesney and Danielle Keats Citron, ‘Deep Fakes: A Looming Challenge for Privacy, Democracy, and National Security’ (2019) 107 California Law Review 1753, coining the term ‘liar’s dividend’ to describe how the mere existence of deepfake technology allows wrongdoers to dismiss authentic evidence as fabricated. https://www.californialawreview.org/print/deep-fakes-a-looming-challenge-for-privacy-democracy-and-national-security/

10. State v Loomis 881 NW2d 749 (Wis 2016), upholding the use of the COMPAS risk-assessment score in sentencing subject to written warnings about the tool’s proprietary and non-transparent methodology. https://www.courtlistener.com/opinion/4237997/state-v-eric-l-loomis/

11. Cf S N Mukherjee v Union of India (1990) 4 SCC 594, on the requirement that administrative and quasi-judicial authorities record reasons in support of their decisions as an incident of natural justice. https://indiankanoon.org/doc/400596/

12. Buckeye Trust v Principal Commissioner of Income Tax, ITA No 1051/Bang/2024, ITAT Bengaluru, order dated 30 December 2024, recalled under section 254(2) of the Income-tax Act 1961 by order dated 7 January 2025 after it was found to rely on three fabricated Supreme Court citations and one fabricated Madras High Court citation. https://itat.gov.in/public/files/upload/1735624480-urq7dV-1-TO.pdf; reported in https://www.livemint.com/money/personal-finance/chatgpt-artificial-intelligence-ai-itat-bengaluru-bench-tax-buckeye-trust-case-errors-11740544685677.html

13. Gummadi Usha Rani v Sure Mallikarjuna Rao, Civil Revision Petition No 2487 of 2025, Andhra Pradesh High Court, judgment dated 21 January 2026 (judicial officer admitted the citations were AI-generated; order upheld because its legal reasoning was sound), reported in https://www.verdictum.in/court-updates/high-courts/andhra-pradesh-high-court/gummadi-usha-rani-v-sure-mallikarjuna-rao-civil-revision-petition-no-2487-of-2025-artificial-intelligence-trial-court-fake-citation-1605575; SLP (C) No 7575 of 2026, Supreme Court of India, order dated 27 February 2026 (reliance on AI-generated fake judgments may amount to misconduct), https://api.sci.gov.in/supremecourt/2026/6571/6571_2026_6_22_68896_Order_27-Feb-2026.pdf; see also https://www.livelaw.in/top-stories/supreme-court-flags-trial-court-citing-ai-generated-fake-judgments-terms-it-misconduct-525097

14. High Court of Kerala, ‘Policy Regarding Use of Artificial Intelligence Tools in the District Judiciary’ (19 July 2025), restricting judicial officers from using AI tools to arrive at findings or draft orders and cautioning against inputting case data into public cloud-based AI tools. https://images.assettype.com/theleaflet/2025-07-22/mt4bw6n7/Kerala_HC_AI_Guidelines.pdf

15. Regulation (EU) 2024/1689 (Artificial Intelligence Act), art 50, which requires deployers of AI systems that generate or manipulate image, audio, or video content constituting a deepfake to disclose that the content has been artificially generated or manipulated, with transparency obligations applicable from 2 August 2026. https://eur-lex.europa.eu/eli/reg/2024/1689/oj

16. Ministry of Electronics and Information Technology, Government of India, public consultations on a proposed Digital India Act to replace the Information Technology Act 2000, addressing emerging technologies including AI (2023, ongoing). https://www.meity.gov.in/writereaddata/files/DIA_Presentation%2009.03.2023%20Final.pdf

17. NITI Aayog, ‘Approach Document for India Part 1 – Principles for Responsible AI’ (February 2021). https://www.niti.gov.in/sites/default/files/2021-02/Responsible-AI-22022021.pdf

18. Bharatiya Nagarik Suraksha Sanhita 2023, s 176(3), which mandates the involvement of forensic experts in the investigation of offences punishable with imprisonment of seven years or more, offered here by analogy for the treatment of AI-generated evidence. https://indiacode.ecourtsindia.com/bnss/section/176/

19. Bharatiya Sakshya Adhiniyam 2023, ss 78–120 (the presumptions chapter of the Adhiniyam), illustrating the existing statutory architecture within which a calibrated presumption against unauthenticated synthetic media could be situated. https://www.indiacode.nic.in/handle/123456789/20063