AI-Generated Evidence in Indian Courts: Is the Law Ready?

AI-Generated Evidence in Indian Courts: Is the Law Ready?

Abstract

Artificial intelligence has quietly entered the courtroom. From deepfake videos
submitted as evidence to AI-enhanced CCTV footage used in criminal trials, the Indian
judiciary is slowly confronting a reality it was not quite built for. This paper examines
whether the existing legal framework in India is equipped to deal with AI-generated
evidence. It looks at the Bharatiya Sakshya Adhiniyam, 2023, the Information Technology
Act, 2000, and relevant constitutional provisions to assess whether AI-generated evidence
can be admitted, verified, and fairly evaluated under current law. The paper argues that while
Indian law has made meaningful progress in recognising digital evidence, it still lacks
specific mechanisms to deal with the unique challenges that AI-generated content brings.
Issues of authenticity, algorithmic bias, explainability, and the risk of deepfake manipulation
remain largely unaddressed. Through a comparison with the United States, European Union,
and United Kingdom, this paper identifies gaps in the Indian framework and suggests
targeted reforms including mandatory forensic verification, judicial training, and the creation
of an AI evidence standard. The central finding is that AI should assist the justice system —
not confuse or overwhelm it.

Keywords:

Artificial Intelligence, Electronic Evidence, Digital Evidence,
Bharatiya Sakshya, Adhiniyam 2023, Deepfakes, Indian Judiciary, AI Ethics

1. Introduction

Not long ago, a judge’s most complex evidentiary challenge might have been deciding
whether a blurry photograph was credible. Today, courts in India and around the world are
beginning to encounter evidence that was not created by any human hand. AI-generated
evidence — images, audio clips, videos, documents, and analytical reports produced or
significantly altered by artificial intelligence — is entering litigation in ways that few legal
systems anticipated.

Artificial intelligence, in simple terms, refers to computer systems capable of
performing tasks that typically require human intelligence: recognising patterns, making
decisions, generating text or visuals, and even predicting outcomes. What makes modern AI
different from earlier software is that it does not just follow fixed rules. It learns from data
and produces outputs that can be indistinguishable from human-created content. A deepfake
video, for instance, can place words in a person’s mouth that they never spoke. An AI image
generator can produce a photograph of an event that never happened. An AI voice cloning
tool can replicate someone’s voice with stunning accuracy.

These capabilities have real implications for the courtroom. Evidence has always
been about establishing truth. The entire structure of evidence law — rules about authenticity,
reliability, hearsay, and burden of proof — exists because human beings and institutions are
fallible, and courts need a disciplined way to distinguish fact from fiction. AI disrupts this
structure in ways that existing law was simply not designed to handle.

India is at a particularly important moment. The country has recently overhauled its
core criminal laws, replacing the Indian Evidence Act, 1872 with the Bharatiya Sakshya
Adhiniyam, 2023. This new legislation updates the treatment of electronic evidence and
reflects a more digital-aware approach to proof. But digital evidence and AI-generated
evidence are not the same thing. A WhatsApp message is digital evidence. A video that has
been AI-manipulated to alter what was said is something else entirely — and the legal tools
to handle that distinction are not yet fully in place.

At the same time, AI is also being used constructively in the justice system. Predictive
policing tools, AI-assisted legal research, sentiment analysis, and even judicial decisionsupport systems are becoming more common. The question is not whether AI belongs in the legal system — it clearly already does — but whether the legal system has the frameworks
to manage it responsibly.

This paper focuses on a specific and urgent problem: when AI-generated content is
submitted as evidence in Indian courts, can the law adequately assess its admissibility, verify
its authenticity, and protect against its misuse? The research problem, simply stated, is this
— Indian evidence law was not written with AI in mind, and the gap between what
technology can now produce and what law can currently regulate is growing wider every
year.

2. Research Objectives

This paper aims to:
(i) Understand what AI-generated evidence is and how it differs from conventional digital
evidence.
(ii) Examine whether AI-generated evidence is admissible under the current Indian legal
framework.
(iii) Analyse the challenges of authenticity, reliability, and explainability that such evidence
raises.
(iv) Compare India’s approach with that of the United States, European Union, and United
Kingdom.
(v) Suggest concrete legal and institutional reforms.

3. Research Questions

(i) What constitutes AI-generated evidence, and how should it be categorised?
(ii) Is AI-generated evidence admissible under the Bharatiya Sakshya Adhiniyam, 2023?
(iii) Does existing Indian law adequately regulate authenticity and reliability of AI evidence?
(iv) What legal reforms are necessary to address the gaps?

4. Research Methodology

This research is doctrinal in nature. It relies on primary sources including the
Constitution of India, statutes such as the Bharatiya Sakshya Adhiniyam, 2023 and the
Information Technology Act, 2000, and judicial decisions of the Supreme Court of India.
Secondary sources include academic books, journal articles, government reports including
those of NITI Aayog and the Law Commission of India, and online legal databases.

5. Meaning of AI-Generated Evidence

AI-generated evidence refers to any material produced, enhanced, or significantly
altered through artificial intelligence and submitted in legal proceedings to prove or disprove
a fact. It is broader than traditional electronic evidence because it is not merely stored or
transmitted by a machine — it is created or transformed by one.

The main types include:
Deepfake videos and images: AI-manipulated audio-visual content that can
convincingly portray events that never occurred or words never spoken. AI-generated voice
recordings: Cloned audio produced to impersonate individuals. AI chat logs: Conversations
generated or interpolated by AI systems. AI-created documents: Reports, contracts, or
records drafted autonomously by language models. AI-enhanced CCTV footage: Lowresolution footage improved through AI upscaling, which may introduce artefacts or alter details. Predictive AI reports: Algorithmic assessments used in bail decisions, risk profiling,
or sentencing recommendations.

Each type carries distinct risks. A deepfake video is a fabrication risk. An AIenhanced CCTV clip is an alteration risk. A predictive report is a bias and transparency risk. The law needs to be sensitive to these differences.

6. Current Legal Framework in India

A. Constitutional Provisions

Article 14 guarantees equality before the law and equal protection of laws. In the
context of AI evidence, this becomes significant when algorithmic tools produce biased
outputs that affect parties unequally — for example, a bail-risk algorithm that
disproportionately flags individuals from certain communities. Article 20(3) protects an
accused from being compelled to be a witness against himself. The question of whether
compelling someone to provide biometric data for AI analysis violates this right remains
unsettled. Article 21, which protects the right to life and personal liberty, has been interpreted
by the Supreme Court to include the right to a fair trial and the right to privacy.
AI-generated evidence that is fabricated, biased, or opaque threatens both.

B. Bharatiya Sakshya Adhiniyam, 2023

The BSA, 2023 is the most relevant statute for this discussion. It replaces the Indian
Evidence Act, 1872 and modernises the treatment of electronic evidence. Under the BSA,
electronic records are admissible as evidence. The Act recognises primary and secondary
electronic evidence, and sets out conditions for admissibility including authentication
through certificates. However, the BSA does not specifically define or address AI-generated
evidence. When an AI has produced or manipulated content, it does not fit neatly into existing
categories. The Act’s provisions assume that an electronic record has a traceable human
author or originating system. AI complicates this assumption. A deepfake video technically
originates from a machine process that may span multiple steps, training datasets, and
generative models — none of which the Act contemplates.

Presumptions under the BSA regarding electronic records functioning in a normal
condition are also problematic when applied to AI evidence, since the ‘normal condition’ of
a generative AI system is precisely to produce novel, realistic-seeming content that may not
correspond to reality.

C. Information Technology Act, 2000

The IT Act deals with electronic records, digital signatures, and various cyber
offences. Section 65B of the earlier Evidence Act — now substantially carried into the BSA
— required a certificate to authenticate electronic records. The IT Act also penalises certain
forms of digital fraud and impersonation, which could theoretically apply to deliberate
deepfake fabrication. However, the IT Act does not specifically address AI-generated
content, and there is no provision dealing with the evidentiary challenges of algorithmic
outputs.

D. Bharatiya Nagarik Suraksha Sanhita, 2023

The BNSS, 2023 introduces provisions for digital and electronic investigation,
including the recording of statements and the collection of electronic evidence during police
investigations. While this represents a step forward, there is no specific guidance on how
investigators should handle AI-generated content encountered during investigations or how
courts should evaluate AI-processed material submitted as part of the chargesheet.

7. Judicial Approach

Indian courts have engaged with electronic evidence through several landmark
decisions. In Anvar P.V. v. P.K. Basheer (2014), the Supreme Court held that electronic
records must comply with the certificate requirement under Section 65B to be admissible as
secondary evidence. This ruling established a strict procedural standard for digital evidence.
In Arjun Panditrao Khotkar v. Kailash Kushanrao Gorantyal (2020), the Court clarified the
conditions under which Section 65B certificates are required and addressed situations where
the certificate cannot be obtained. These cases establish an important precedent: courts take
authentication of electronic evidence seriously.
Justice K.S. Puttaswamy (Retd.) v. Union of India (2017) is significant for a different
reason. The nine-judge bench unanimously recognised privacy as a fundamental right under
Article 21. This has direct relevance to AI evidence — the mass collection of biometric or
personal data to train or deploy AI systems in criminal investigations raises serious privacy
concerns that courts will eventually need to confront.

What these cases do not do is provide a framework specifically for AI-generated
evidence. The existing judicial approach focuses on whether electronic records were properly
certified and stored. It does not ask whether the content itself might have been algorithmically
fabricated, whether the AI system used was reliable, or whether the model was biased. These
are the questions that the next generation of evidence jurisprudence will need to answer.

8. Challenges of AI-Generated Evidence

A. Authenticity

Proving that an AI output is genuine and unaltered is far more complex than verifying
a traditional document. AI systems can produce highly convincing fabrications, and standard
forensic tools may not be equipped to detect them. The party relying on AI evidence must be
able to show that the system used was reliable, properly calibrated, and that the output has
not been tampered with.

B. Deepfake Technology

Deepfakes represent perhaps the most serious evidentiary threat. A fabricated video
can be produced in hours using freely available software and submitted as evidence in
criminal proceedings. Without mandatory forensic verification, courts have no reliable way
to detect such fabrications. The consequences — wrongful conviction, acquittal of the guilty,
reputational destruction — are severe.

C. Chain of Custody

Traditional evidence law requires a clear chain of custody — a documented record of
who handled evidence and when. AI complicates this because the ‘evidence’ may have passed
through multiple systems, been processed by algorithms, and been stored in forms that are
difficult to trace. Each step in the AI pipeline is a potential point of alteration.

D. Explainability and the Black-Box Problem

Many AI systems, particularly those based on deep learning, function as black boxes.
They produce outputs without providing any explanation of the reasoning behind those
outputs. Judges cannot evaluate what they cannot understand. If an AI system flags a person
as a flight risk, or identifies a suspect from surveillance footage, but cannot explain how it
arrived at that conclusion, then the court is being asked to trust a machine it cannot scrutinise
— which is fundamentally incompatible with the principles of evidence law.

E. Algorithmic Bias

AI systems learn from historical data. If that data reflects existing social biases —
racial, economic, gender-based — the system will reproduce and potentially amplify those
biases in its outputs. A facial recognition system trained predominantly on one demographic
group may perform poorly on others, leading to misidentification. In a criminal trial, this is
not just a technical error — it is a justice failure.

F. Privacy and Data Protection

The collection of personal data to train or operate AI systems raises fundamental
privacy concerns, particularly given the Supreme Court’s recognition of privacy as a
fundamental right. Investigative AI tools that harvest large volumes of personal data without
consent or legal authorisation may violate constitutional guarantees.

G. Burden of Proof

Under current law, it is unclear who bears the burden of proving that AI-generated
evidence is reliable. The proponent of the evidence should logically bear this burden, but
existing procedural rules are not specific on this point. Nor is there clarity on what standard
of proof — or what type of expert testimony — is required.

9. Comparative Analysis

A. United States

The Federal Rules of Evidence in the United States apply to AI-generated evidence
through general provisions on authentication and reliability. Courts have begun to grapple
with AI evidence in various contexts, including facial recognition and predictive risk
assessments. Some states have enacted specific legislation on deepfakes, particularly in the
context of elections and non-consensual intimate imagery. The Daubert standard, which
requires scientific evidence to be based on sufficient facts and reliable methodology, provides
courts with a framework to scrutinise algorithmic evidence — though its application to AI
remains inconsistent.

B. European Union

The EU AI Act, which came into force in 2024, establishes a risk-based framework
for AI systems. High-risk AI applications — including those used in law enforcement and
the administration of justice — are subject to strict requirements for transparency, accuracy,
and human oversight. The GDPR provides strong protections for personal data, including
data used to train AI systems. Together, these instruments create a more comprehensive
regulatory environment for AI in legal proceedings than currently exists in India.

C. United Kingdom

The UK has relied primarily on existing evidence rules and judicial discretion to
manage digital and AI-generated evidence. Courts retain broad powers to exclude evidence
that is unreliable or whose admission would be unfair. The Law Commission of England and
Wales has undertaken a review of the law on automated processes and evidence, recognising
the need for clearer rules. The UK approach is more flexible but also less certain — there is
no dedicated statutory framework for AI evidence.

10. Need for Legal Reforms

India does not yet have AI-specific evidentiary rules. The BSA, 2023 is an
improvement over the 1872 Act, but it remains anchored to a model of evidence that assumes
human authorship and traceable provenance. Several reforms are urgently needed.

There is a need for dedicated AI evidence rules that distinguish between AI-assisted
evidence (where AI aids human-generated content) and AI-generated evidence (where the
output is substantially the product of an algorithm). Authentication standards must be
updated to require forensic verification of AI-generated content before it is admitted. Expert
witness requirements should be clarified so that courts have access to qualified AI forensics
experts when evaluating such evidence. Judicial and legal education on AI is essential —
judges cannot evaluate what they do not understand. National AI forensic laboratories should
be established to provide independent verification services. A statutory certification
mechanism for AI tools used in law enforcement would bring transparency and
accountability. Finally, India should consider an AI ethics framework for legal AI, governing
how AI systems deployed in investigations and courts should be designed, audited, and held
accountable.

11. Critical Analysis

Looking at all of this honestly, the conclusion is uncomfortable but necessary: Indian
evidence law is not ready for AI-generated evidence. The BSA, 2023 is a genuine
improvement over the old Evidence Act in its treatment of electronic records. But updating
the law to handle digital evidence is not the same as preparing it for AI evidence. The
difference matters enormously.
A digital photograph taken on a smartphone is electronic evidence. The same
photograph, manipulated by a deepfake algorithm to alter a person’s expression or
superimpose someone else’s face, is AI-generated evidence — and the legal tools needed to
detect, challenge, and evaluate it are simply not the same tools.
The existing framework’s reliance on Section 65B certificates assumes that the
authenticity question is primarily one of whether a record was properly extracted and
transmitted. But with AI-generated content, the authenticity question runs deeper: was the
content itself created or altered by a machine? No certificate addresses that. No existing
presumption in the BSA accounts for it.
The black-box problem deserves particular emphasis. A cornerstone of fair trial rights
is the ability of the accused to challenge the evidence against them. If a court accepts an AI
system’s output — say, a facial match from CCTV footage — without the accused being able
to understand or challenge the algorithm’s logic, this is not just a technical problem. It is a
due process problem. Courts that accept opaque AI outputs as reliable evidence are, in effect,
outsourcing judicial reasoning to machines they cannot hold accountable.

AI should be a tool that assists judges — by processing large volumes of data,
identifying patterns, or enhancing evidence — not a substitute for judicial reasoning. The
moment a court defers to an AI output without subjecting it to the same scrutiny as any other
piece of evidence, something important is lost.

12. Suggestions

1. AI-specific evidentiary rules should be incorporated into the BSA or enacted through
a separatestatute, distinguishing between AI-assisted and AI-generated evidence and
establishing clear admissibility standards.
2. Mandatory forensic verification should be required before any AI-generated content
is admitted asevidence, with the burden placed on the party relying on such evidence.
3. Digital watermarking and provenance tracking standards should be established so
thatAI-generated content can be traced to its origin system.
4. National AI forensic laboratories should be created under a central or state authority
to provideindependent, scientifically valid verification of AI-generated evidence.
5. Judicial guidelines on AI evidence should be issued by the Supreme Court or High
Courts,addressing how such evidence should be evaluated, what expert testimony is required,
and what disclosure obligations apply.
6. Training programmes for judges, prosecutors, defence lawyers, and police officers
should beintroduced, covering the basics of AI systems, their limitations, and the risks of
algorithmic bias.
7. An AI ethics framework for legal AI should be developed, setting standards for
fairness,transparency, and accountability in AI systems used in investigations, bail decisions,
and other justice-related contexts.
8. Stronger cyber laws should address the deliberate creation and submission of
deepfakeevidence, with specific criminal penalties for AI-based evidence fraud.

13. Conclusion

There is something worth pausing on before concluding. The Indian legal system has,
over the past century and a half, navigated enormous technological change. Photograph
evidence, tape recordings, electronic records — each time, there was a period of legal uncertainty before the law caught up. AI-generated evidence is the next such challenge, and in some ways it is the most difficult one yet.

AI is not just a new medium for storing or transmitting information. It is a system
capable of creating information — and of doing so in ways that are indistinguishable from
reality. That is a qualitatively different challenge from anything evidence law has previously
faced. The BSA, 2023 represents progress. It is cleaner and more comprehensive in its
treatment of digital evidence than its predecessor. But it was not written with AI in mind, and
the gap between what AI can produce and what the law can currently evaluate is not a minor
technical gap. It is a substantial one that will only grow as AI capabilities advance.
Courts across India will increasingly encounter AI-generated content — in family
disputes where AI-generated messages are submitted, in criminal trials where deepfake
videos appear, in corporate litigation where AI-produced documents are tendered. In each
case, the judge will need tools that the current framework does not fully provide: tools to
assess authenticity, to demand explanation, to evaluate bias, and to protect the fundamental
fairness of the trial.
The comparative experience of the United States, the European Union, and the United
Kingdom offers useful lessons. Risk-based AI regulation, transparency requirements, and
clear forensic standards for evidentiary AI use are not aspirational — they are already being
implemented in peer jurisdictions. India must move in the same direction, ideally faster.
The aim of evidence law has always been truth — not perfect truth, which is beyond
any legal system’s reach, but disciplined, reasoned, procedurally fair truth. AI presents both
a threat and an opportunity in that pursuit. Managed well, it can help courts process evidence
more efficiently and identify patterns that human analysis might miss. Managed poorly, it
can introduce new forms of error, bias, and manipulation that undermine justice at a
fundamental level.
India has the legislative tools, the judicial tradition, and the institutional capacity to
get this right. What is needed now is will — the political and judicial will to recognise that
AI-generated evidence is a distinct legal category requiring distinct legal treatment, and to
act on that recognition before the courts are overwhelmed by problems that could have been
anticipated and addressed.

The law must not be perpetually catching up with technology. For once, it should try
to get ahead of it.

14. References

A. Primary Sources
(i) Constitutional Provisions
Constitution of India, 1950 — Articles 14, 20(3), 21.
(ii) Statutes
1. Bharatiya Sakshya Adhiniyam, 2023.
2. Bharatiya Nagarik Suraksha Sanhita, 2023.
3. Information Technology Act, 2000.
4. Indian Evidence Act, 1872 (repealed).
(iii) Judicial Decisions
1. Anvar P.V. v. P.K. Basheer, (2014) 10 SCC 473.
2. Arjun Panditrao Khotkar v. Kailash Kushanrao Gorantyal, (2020) 7 SCC 1.
3. Justice K.S. Puttaswamy (Retd.) v. Union of India, (2017) 10 SCC 1.

B. Secondary Sources
(i) Books
1. Woodrow Barfield and Ugo Pagallo (eds.), Research Handbook on the Law of
ArtificialIntelligence (Edward Elgar, 2018).
2. Stephen Mason (ed.), Electronic Evidence (4th edn, Institute of Advanced Legal Studies,
2017).
3. Ryan Calo, Artificial Intelligence Policy: A Primer and Roadmap (2017).
(ii) Journal Articles
1. Vaibhav Parikh and Gowree Gokhale, ‘Admissibility of Electronic Evidence in India’
(2019) IndianJournal of Law and Technology.
2. Ryan Abbott, ‘I Think, Therefore I Invent: Creative Computers and the Future of
Patent Law'(2016) 57 Boston College Law Review 1079.

(iii) Government Reports
1. NITI Aayog, National Strategy for Artificial Intelligence (2018).
2. NITI Aayog, Responsible AI for All (2021).
3. Law Commission of India, Report No. 200 (2006).
(iv) International Legislation and Instruments
1. European Parliament and Council, EU Artificial Intelligence Act, 2024.
2. General Data Protection Regulation (GDPR), Regulation (EU) 2016/679.
3. Federal Rules of Evidence (United States), 28 U.S.C. App.