AI-GENERATED CONTENT AND DEEPFAKE EVIDENCE: THE FUTURE OF ADMISSIBILITY, AUTHENTICATION AND PROOF IN INDIAN COURTS

Abstract

The rapid evolution of Generative Artificial Intelligence has fundamentally altered the nature of digital information. Images, videos, voice recordings, documents and even apparently authentic human communications can now be generated or manipulated by artificial intelligence with an unprecedented degree of realism. Deepfake technology, synthetic media and AI-generated content consequently pose a serious challenge to the traditional assumptions upon which courts have historically relied while assessing documentary and electronic evidence.

The central difficulty is no longer merely whether an electronic record exists, but whether the record accurately represents the underlying event, whether the person depicted actually performed the act attributed to him or her, whether the recording has been manipulated, and whether the provenance and chain of custody of the digital material can be satisfactorily established.

Indian evidence law has already recognised the legal significance of electronic records. The Bharatiya Sakshya Adhiniyam, 2023 (“BSA”) expressly provides that an electronic or digital record shall not be denied legal effect merely because it is electronic or digital and prescribes a specific statutory framework for proving such records. However, the emergence of sophisticated AI-generated media raises questions which conventional electronic-evidence doctrine was not specifically designed to answer.

This article examines the evidentiary challenges posed by AI-generated content and deepfakes, analyses the existing Indian legal framework, considers the principles of authenticity, reliability, provenance and chain of custody, and proposes a future-oriented judicial methodology for dealing with synthetic evidence without compromising either technological progress or the fundamental requirements of a fair trial.

Keywords: Artificial Intelligence, Generative AI, Deepfake, Synthetic Media, Electronic Evidence, Digital Evidence, Authentication, Chain of Custody, Bharatiya Sakshya Adhiniyam, 2023, Cyber Forensics, Digital Forensics, Admissibility, Evidentiary Reliability.

1. Introduction

The law of evidence has traditionally operated on a relatively simple premise: a photograph represents something that was photographed; a video recording represents an event captured by a camera; an audio recording represents words spoken by a person; and a document represents information recorded by a human or an identifiable institutional process.

Artificial Intelligence is rapidly weakening these assumptions.

A modern generative-AI system can create a photorealistic image of a person who never occupied the depicted location. It can reproduce a person’s voice and generate speech that the person never uttered. It can manipulate facial expressions, lip movements and body movements in an existing video. It can create apparently genuine documents, emails, photographs and other digital artefacts.

The evidentiary problem therefore moves beyond the traditional question:

“Is this electronic record genuine?”

The more difficult question is:

“What exactly is genuine about this electronic record?”

A digital file may be genuine in the sense that it exists on a particular device, yet the content contained in that file may be synthetically generated or manipulated. Conversely, a file may have been compressed, edited for legitimate purposes or converted between formats without necessarily becoming false.

The distinction between digital authenticity, content authenticity, provenance, and truthfulness of the underlying event will therefore become increasingly important.

NIST’s recent work on digital-media forensics similarly emphasises provenance, authentication and the distinction between synthetic, repurposed and traditionally manipulated media.

The future courtroom may consequently encounter a new class of evidentiary disputes in which the parties do not merely disagree about what a witness says, but about whether the digital reality itself can be trusted.

2. Understanding AI-Generated Content and Deepfakes

2.1 Artificial Intelligence Generated Content

AI-generated content refers broadly to material produced wholly or substantially through artificial intelligence systems.

It may include:

AI-generated photographs;
AI-generated videos;
synthetic speech;
cloned voices;
AI-generated documents;
manipulated CCTV footage;
synthetic social-media conversations;
AI-generated emails;
digitally reconstructed events;
face-swapped videos;
digitally altered biometric material; and
AI-generated text purportedly written by a particular person.

Not all AI-generated material is unlawful or deceptive. AI may legitimately be used for artistic, educational, commercial, scientific or investigative purposes.

The legal difficulty arises when synthetic content is presented as an authentic representation of a real-world event.

2.2 Deepfakes

A deepfake is a form of synthetic or manipulated digital media in which artificial intelligence techniques are used to create or alter an image, audio recording or video so that it appears to depict a person, statement or event that may not have occurred in reality.

Deepfakes can involve:

Face swapping;
Voice cloning;
Facial reenactment;
Lip-sync manipulation;
Synthetic video generation;
Context manipulation;
AI-generated persons; and
Combination of genuine and synthetic material.

The evidentiary danger is particularly acute because the human eye and ear can no longer reliably distinguish authentic media from sophisticated synthetic media.

NIST has specifically noted that modern generative AI makes highly realistic deepfakes increasingly accessible and that detection systems can suffer substantial degradation when moved from controlled research environments to real-world forensic conditions.

3. The Transformation of the Traditional Evidentiary Paradigm

Traditional evidence law generally distinguishes between:

oral evidence;
documentary evidence;
primary evidence;
secondary evidence;
expert evidence; and
circumstantial evidence.

Digital evidence introduced a new technological dimension to this classification.

AI-generated evidence creates an additional layer.

Consider the following hypothetical:

A CCTV recording allegedly shows an accused committing an offence at 10:00 p.m.

The prosecution produces:

the CCTV file;
a copy on a pen drive;
screenshots;
a transcript;
metadata; and
a forensic report.

The defence alleges that the video is an AI-generated face-swap.

The issue is no longer simply whether the CCTV system produced the file.

The court must potentially determine:

whether the recording originated from the claimed camera;
whether the original recording still exists;
whether the file was altered;
whether the facial image was substituted;
whether the audio was synthetically generated;
whether metadata is reliable;
whether the storage system was compromised;
whether the forensic examination was scientifically valid;
whether the AI detector used by the expert is reliable; and
whether the accused can be identified with sufficient certainty.

This represents a fundamental shift from electronic admissibility to algorithmic and forensic authenticity.

4. The Indian Legal Framework

4.1 Bharatiya Sakshya Adhiniyam, 2023

The Bharatiya Sakshya Adhiniyam, 2023 represents India’s current statutory framework governing evidence.

Section 61 provides that an electronic or digital record cannot be denied admissibility merely because it is electronic or digital and gives such records the same legal effect, validity and enforceability as other documents, subject to the statutory requirements. Section 62 specifically deals with proof of the contents of electronic records, while Section 63 prescribes the special framework concerning electronic records.

The significance of these provisions is substantial.

The law does not treat electronic material as inherently inferior merely because it exists in digital form.

However:

Admissibility of an electronic record does not automatically establish the truthfulness, integrity or reliability of its contents.

This distinction becomes particularly important in the context of deepfakes.

5. From Section 65B to Section 63 of the BSA: The Evolution of Electronic Evidence

Under the Indian Evidence Act, 1872, Section 65B became the principal statutory provision concerning electronic evidence.

The Supreme Court’s decisions in Anvar P.V. v. P.K. Basheer, (2014) 10 SCC 473 and Arjun Panditrao Khotkar v. Kailash Kushanrao Gorantyal, (2020) 7 SCC 1 established important principles regarding the proof of electronic records.

In Anvar P.V., the Supreme Court emphasised the statutory requirements governing electronic records. In Arjun Panditrao, the three-Judge Bench reaffirmed the importance of the statutory certificate requirement for secondary electronic evidence, while also recognising that a certificate is unnecessary where the original electronic record itself is produced in the manner contemplated by law.

The current BSA has replaced the former statutory architecture with its own provisions, including Section 63.

This legislative transition is important because future litigation involving AI-generated evidence will necessarily require courts to apply the principles of electronic-record admissibility to a technologically more complex category of material.

6. The Fundamental Distinction: Admissibility Is Not Proof of Authenticity

This may become one of the most important principles in future AI litigation.

A court should distinguish between four separate questions:

First — Is the material relevant?

Does the material have a logical connection with a fact in issue?

Second — Is the material legally admissible?

Does it satisfy the applicable statutory requirements?

Third — Is the material authentic?

Does the material genuinely originate from the source claimed by the party producing it?

Fourth — Is the content reliable and truthful?

Does the material accurately represent the underlying event?

These questions must not be collapsed into one.

For example, a WhatsApp video may be relevant and may satisfy the procedural requirements for production of an electronic record. That does not automatically establish that the video has not been manipulated.

Similarly, a file may have a valid electronic signature while the factual assertions contained in an accompanying AI-generated narrative remain disputed.

7. Deepfake Evidence and the Problem of Authentication

Authentication is likely to become the central evidentiary battleground.

The party relying upon a digital recording may need to establish:

the source of the recording;
the device used for creation;
the date and time of creation;
the original storage location;
the chain of custody;
the method of extraction;
the integrity of the file;
whether any editing occurred;
whether compression occurred;
whether metadata was altered;
whether the file passed through social-media platforms;
whether the device itself was compromised; and
whether forensic examination identifies evidence of synthetic generation.

NIST’s guidance on digital media emphasises the importance of determining provenance and understanding the lifecycle of the media, including capture, transfer, editing, copying, compression and storage.

Accordingly, the future forensic inquiry should not be limited to:

“Is this a deepfake?”

It should also ask:

“What is the provenance of this file, and what happened to it from the moment of capture until its production before the Court?”

8. The Problem with AI Detection Tools

An obvious solution would appear to be the use of AI itself to detect AI-generated content.

However, this creates a paradox.

If AI creates a deepfake, can another AI conclusively establish that the first AI created it?

The answer, in many circumstances, may be no.

Deepfake detection systems may produce:

false positives;
false negatives;
inconsistent results;
model-dependent results;
results affected by compression;
results affected by resizing;
results affected by subsequent editing; and
results which become unreliable against newer generation models.

NIST’s current deepfake-forensics work specifically identifies challenges concerning generalisation, post-processing and anti-forensic techniques.

Therefore, a forensic expert should not simply state:

“The AI detector says this is a deepfake.”

The expert should explain:

what methodology was used;
what dataset or benchmark supports that methodology;
the known error rate;
whether the tool has been independently validated;
whether the relevant media was compressed;
whether the algorithm was tested against comparable generation models;
whether alternative explanations exist; and
the degree of scientific confidence in the conclusion.

9. Expert Evidence Will Assume Greater Importance

The future courtroom is likely to witness an increased dependence upon experts in:

digital forensics;
cyber forensics;
audio forensics;
video forensics;
image authentication;
biometric analysis;
machine learning;
computer science;
metadata analysis; and
provenance verification.

However, expert evidence should assist the Court; it should not replace the judicial function.

An expert should ideally distinguish between:

Observation

“What the forensic examination revealed.”

Inference

“What those findings may indicate.”

Conclusion

“The degree to which the evidence supports authenticity or manipulation.”

Limitation

“What the forensic methodology cannot establish.”

This distinction is particularly important because the technological sophistication of AI may create an illusion of scientific certainty.

10. Voice Cloning and AI-Generated Audio

Voice cloning may become one of the most dangerous forms of synthetic evidence.

A recording may apparently contain the voice of:

an accused;
a complainant;
a witness;
a public official;
a business executive; or
a family member.

Suppose a purported recording contains a confession.

The defence claims that the accused’s voice was cloned through AI.

The Court may then have to examine:

the original recording;
the recording device;
the source file;
background noise;
acoustic characteristics;
compression;
waveform characteristics;
speaker identification;
possible synthetic-generation indicators;
chain of custody; and
independent forensic examination.

A voice sample alone may therefore become insufficient where sophisticated cloning technology is reasonably alleged.

11. AI-Generated Video and Facial Identification

Facial recognition technology creates another serious challenge.

A face appearing in a video does not necessarily establish that the person actually participated in the depicted event.

A deepfake can preserve:

the body of one individual;
the environment of another event; and
the face of a third individual.

Consequently, identification should increasingly involve a broader forensic assessment of:

body morphology;
gait;
clothing;
environmental consistency;
shadows and reflections;
lighting;
temporal continuity;
facial movements;
audio synchronisation;
source footage; and
surrounding digital evidence.

The evidentiary value of facial identification should therefore be evaluated in the context of the complete evidentiary record rather than through visual resemblance alone.

12. The Chain of Custody Problem

Chain of custody is not a mere procedural formality in deepfake litigation.

Every transfer of digital evidence may create an opportunity for:

modification;
deletion;
compression;
metadata alteration;
format conversion;
overwriting; or
substitution.

Accordingly, investigators should preserve the original material and document:

Capture → Preservation → Acquisition → Hashing → Examination → Storage → Transfer → Production

Where possible, cryptographic hashes should be generated and preserved at appropriate stages.

A forensic record should identify:

who seized the device;
when it was seized;
how it was stored;
who accessed it;
what forensic tools were used;
what extraction methodology was adopted;
whether the original device was altered; and
whether subsequent copies correspond to the original hash.

NIST has recognised that preservation of digital evidence presents unique challenges beyond traditional evidence preservation.

13. The Evidentiary Value of Metadata

Metadata may assist in determining:

creation time;
modification time;
device information;
software used;
file format;
location information;
encoding history; and
other technical characteristics.

However, metadata should not be treated as infallible.

Metadata can sometimes be:

stripped;
rewritten;
altered;
lost during transmission;
changed during editing; or
affected by platform processing.

Thus, metadata should ordinarily form part of a broader authenticity assessment rather than constitute conclusive proof by itself.

14. Social Media and Deepfake Evidence

Social-media platforms have become major distribution channels for synthetic media.

A deepfake may travel through:

AI Tool → Device → Messaging App → Social Media → Download → Screenshot → Re-upload → Court

Each stage may modify the evidentiary characteristics of the original material.

A screenshot is therefore particularly problematic.

It may demonstrate that a particular representation appeared on a screen, but it may not independently establish:

who created the content;
whether the original content was authentic;
whether the screenshot reflects the complete context;
whether the underlying file existed in the claimed form; or
whether the content had previously been manipulated.

Courts should therefore be cautious in treating screenshots as conclusive proof of the underlying digital event.

15. AI-Generated Documents and Fabricated Communications

The problem is not confined to photographs and videos.

Generative AI can create apparently authentic:

emails;
invoices;
letters;
contracts;
legal notices;
social-media conversations;
chat messages;
screenshots;
corporate communications; and
internal memoranda.

In commercial litigation, an AI-generated email allegedly sent by a director could potentially become the subject matter of a dispute concerning:

authorship;
intention;
authority;
execution;
electronic signature;
server logs;
account access;
IP addresses;
device history; and
metadata.

The mere appearance of a document should therefore become increasingly insufficient where its provenance is disputed.

16. AI-Generated Legal Material: A Different but Related Problem

Artificial intelligence can also generate fictitious case citations, fabricated quotations and non-existent judicial decisions.

This is no longer a purely theoretical concern.

In Pooja Ramesh Singh v. Jammu and Kashmir Bank Ltd., 2026 INSC 668, the Supreme Court addressed the consequences of courts and tribunals relying upon AI-generated or hallucinated case law and emphasised the importance of verifying legal authorities.

Although this is not a deepfake-media case, it illustrates a broader principle:

AI-generated information cannot acquire legal authority merely because it appears plausible or is generated by an apparently sophisticated technological system.

The same principle should apply to synthetic photographs, videos, audio and documents.

17. Constitutional Dimensions

Deepfake litigation will inevitably engage constitutional principles.

Article 14 — Equality and Fairness

A judicial process based upon unreliable synthetic evidence may undermine substantive fairness.

Article 19 — Freedom of Speech and Expression

Not every AI-generated representation is unlawful.

Satire, parody, artistic expression, political commentary and creative expression may raise legitimate Article 19 questions.

The law must therefore distinguish between:

expression;
misinformation;
deception;
defamation;
impersonation;
fraud; and
unlawful manipulation.

Article 21 — Life and Personal Liberty

Deepfakes can affect:

dignity;
reputation;
privacy;
bodily autonomy;
personal security; and
fair trial rights.

In criminal proceedings, the use of fabricated digital evidence can potentially threaten the liberty of an accused in the most direct manner.

18. Presumption of Innocence and Deepfake Evidence

Criminal jurisprudence operates upon the foundational principle that the prosecution must establish guilt in accordance with law.

Deepfake technology creates a peculiar danger.

A fabricated video may appear more convincing than a genuine but imperfect eyewitness account.

Human beings are naturally inclined to believe visual evidence.

The courtroom, however, must resist the psychological assumption:

“Seeing is believing.”

In the age of generative AI, the more appropriate principle may become:

“Seeing is only the beginning of verification.”

Where the prosecution relies substantially upon disputed synthetic media, the Court must carefully evaluate authenticity and reliability before attaching decisive evidentiary weight to it.

19. Can a Deepfake Ever Be Admissible?

Yes.

A crucial distinction must be made between “deepfake” as a description of synthetic media and “inadmissible evidence” as a legal conclusion.

Suppose a prosecution produces an AI-generated reconstruction to explain the sequence of an event.

If the reconstruction is:

clearly identified as AI-generated;
not represented as an original recording;
scientifically explained;
relevant to the issue;
supported by underlying evidence; and
used merely as demonstrative material,

its legal treatment may differ fundamentally from a fabricated video falsely presented as an authentic CCTV recording.

Therefore:

Synthetic does not necessarily mean inadmissible; deceptive does not necessarily mean irrelevant; and authentic does not automatically mean truthful.

The purpose for which the material is tendered is critical.

20. The Need for a Multi-Layer Authentication Test

Indian courts may eventually benefit from adopting a structured approach.

A proposed Deepfake Evidence Authentication Test could contain the following stages:

Stage I — Source Verification

Who created or captured the material?

Stage II — Device Verification

From which device did it originate?

Stage III — File Integrity

Does the produced file correspond to the original?

Stage IV — Provenance Analysis

What happened to the file between creation and production?

Stage V — Metadata Examination

What does the available metadata indicate?

Stage VI — Forensic Examination

Are there indicators of manipulation or synthetic generation?

Stage VII — Algorithmic Validation

If an AI detection system is used, what is its reliability and error rate?

Stage VIII — Human Expert Review

Has a qualified independent expert examined the material?

Stage IX — Corroboration

Is the material supported by independent evidence?

Stage X — Contextual Integrity

Does the media accurately represent the relevant event in its complete temporal and contextual setting?

Such a framework would prevent courts from treating a single AI-detection score as determinative.

21. The Need for a Higher Standard Where Digital Evidence Is Dispositive

Not every electronic record requires the same level of forensic scrutiny.

The evidentiary approach should be proportionate to the importance of the material.

For example:

A minor commercial communication may not require the same forensic examination as a video allegedly proving murder.

Similarly, a disputed AI-generated photograph may require more rigorous scrutiny if it is the principal evidence upon which a person’s liberty or reputation depends.

The greater the evidentiary consequence, the greater should be the emphasis on:

authenticity;
provenance;
independent corroboration;
forensic validation; and
procedural fairness.

22. Role of the Trial Judge

The judge will become increasingly important as a technological gatekeeper.

The Court need not become an expert in artificial intelligence.

However, judicial officers dealing with technologically complex evidence should possess sufficient technological literacy to ask appropriate questions.

A Court should be able to ask:

What exactly is this file?
Who created it?
How was it obtained?
What is the original?
What is the hash value?
Has the file been edited?
What forensic methodology was used?
What is the error rate of that methodology?
Can the methodology distinguish new-generation AI from genuine media?
Has the expert examined the original or merely a compressed copy?
Is the expert’s conclusion independently reproducible?
Is there corroborative evidence?

Judicial reasoning should record these considerations wherever the authenticity of digital evidence is seriously contested.

23. The Role of the Defence Counsel

For defence lawyers, AI-generated evidence creates an entirely new area of cross-examination.

A defence counsel may appropriately investigate:

Whether the original file has been produced;
Whether the device was seized;
Whether forensic imaging was performed;
Whether hash values were recorded;
Who had access to the device;
Whether the file was edited;
Whether the investigating agency possesses the complete recording;
Whether the alleged recording has gaps;
Whether metadata is consistent;
Whether the forensic expert tested the latest AI-generation techniques;
Whether the expert’s methodology has been independently validated;
Whether alternative explanations were excluded; and
Whether the alleged digital evidence is corroborated by independent evidence.

The defence should not merely assert:

“This is a deepfake.”

It should identify why the prosecution’s authentication process is inadequate.

24. Role of the Prosecution

The prosecution, on the other hand, should anticipate the deepfake defence.

Whenever important digital evidence is collected, investigators should preserve:

the original device;
original storage media;
forensic images;
hash values;
metadata;
system logs;
relevant server records;
complete recordings;
surrounding footage;
extraction reports; and
chain-of-custody documentation.

The prosecution should preferably avoid relying exclusively upon a single digital artefact where independent corroboration is available.

25. The Emerging Concept of Digital Provenance

One of the most promising solutions is content provenance.

Provenance attempts to establish the history of digital content from creation through subsequent transformations.

Technological approaches may include:

cryptographic hashes;
digital signatures;
trusted capture devices;
tamper-evident logs;
content credentials;
watermarking;
secure metadata;
device attestation; and
cryptographically verifiable provenance systems.

NIST has identified provenance tracking, watermarking, synthetic-content detection and related transparency mechanisms as important approaches to reducing risks associated with synthetic content.

However, provenance technology should be regarded as an aid to authentication, not as an absolute substitute for judicial evaluation.

26. The Future of CCTV Evidence

CCTV evidence will probably become one of the most contested categories of digital evidence.

Courts may increasingly require:

original DVR/NVR data;
system logs;
camera configuration details;
frame-rate information;
timestamp verification;
storage history;
forensic extraction reports;
complete footage rather than selected clips; and
expert analysis where manipulation is alleged.

A short video clip downloaded from a messaging application should not automatically be treated as equivalent to the original footage preserved on the CCTV system.

27. Deepfake Evidence and the Right to Cross-Examination

Where a digital recording is relied upon against an accused, meaningful cross-examination may require access to:

the original file;
forensic report;
metadata;
extraction methodology;
relevant software;
expert methodology; and
underlying material upon which the expert formed the opinion.

If an expert relies upon a proprietary AI system whose methodology is entirely opaque, difficult questions may arise concerning transparency and effective cross-examination.

The adversarial process cannot operate effectively if the party challenging the evidence is unable to understand the basis upon which authenticity is asserted.

28. Privacy and Data Protection Concerns

Deepfake investigations may require access to highly personal data, including:

private photographs;
personal videos;
biometric information;
voice samples;
mobile-phone data;
cloud accounts; and
private communications.

Investigators must therefore balance evidentiary requirements with legal protections concerning privacy and lawful access to personal information.

The forensic examination of a device should remain connected to the legitimate investigative purpose and should not become an unrestricted examination of a person’s entire digital life.

29. The Problem of “Deepfake Denial”

The opposite danger must also be recognised.

Once society becomes aware of deepfake technology, an accused or litigant may simply allege:

“The video is AI-generated.”

That allegation cannot, by itself, destroy otherwise credible evidence.

Otherwise, deepfake technology may become a universal escape mechanism.

Therefore, the law must avoid both extremes:

Extreme One

“Digital evidence is inherently reliable.”

Extreme Two

“Any digital evidence can be dismissed as a deepfake.”

The correct approach lies between these extremes: structured authentication, scientific examination, corroboration and judicial assessment.

30. Future Legislative Reforms

India may eventually require more specific legislative or procedural provisions concerning synthetic evidence.

Possible reforms include:

1. Statutory definition of synthetic evidence

A legal definition could distinguish:

AI-generated content;
partially synthetic content;
manipulated content;
repurposed content; and
authentic digital media.

2. Special forensic protocols

Standard operating procedures may be prescribed for the seizure and examination of suspected deepfake material.

3. Expert accreditation

Courts may benefit from recognised standards for digital-media forensic experts.

4. AI detector validation

AI-based forensic tools should undergo independent validation.

5. Provenance standards

Government agencies, courts and critical institutions may increasingly adopt tamper-evident digital capture systems.

6. Judicial training

Judicial academies may incorporate modules on:

generative AI;
synthetic media;
deepfakes;
digital forensics;
AI hallucinations; and
algorithmic evidence.

7. Investigative training

Police and prosecution agencies require specialised training in preserving and authenticating digital evidence.

31. Comparative and International Perspective

The deepfake problem is not confined to India.

Globally, governments, forensic laboratories and technology institutions are working on:

synthetic-media detection;
provenance systems;
watermarking;
digital signatures;
media authentication;
AI governance; and
forensic benchmarks.

NIST’s ongoing forensic initiatives demonstrate that the international scientific community recognises the need to test deepfake detection systems against realistic, post-processed and adversarial media rather than relying solely upon laboratory conditions.

India can benefit from these developments while developing solutions compatible with its constitutional, procedural and evidentiary framework.

32. A Proposed Judicial Test for Suspected Deepfake Evidence

When a party specifically alleges that an electronic record is AI-generated or manipulated, the Court may consider adopting the following structured questions:

Question 1

What is the original electronic record?

Question 2

Who created or captured it?

Question 3

From which device or system was it obtained?

Question 4

Has the original device or source been preserved?

Question 5

Can the chain of custody be demonstrated?

Question 6

Are the relevant metadata and hash values available?

Question 7

Has the media been edited, compressed, converted or transmitted through a third-party platform?

Question 8

What forensic methodology has been used to determine authenticity?

Question 9

Has that methodology been scientifically validated?

Question 10

What are its known limitations and error rates?

Question 11

Has an independent expert reviewed the findings?

Question 12

Is the alleged digital evidence corroborated by independent evidence?

Question 13

What is the consequence of accepting or rejecting the digital material?

Question 14

Would reliance upon the disputed material cause prejudice to a party’s substantive rights?

This approach could assist courts in maintaining a disciplined evidentiary analysis.

33. The Future Courtroom: From Digital Evidence to Computational Evidence

The next generation of litigation may involve evidence that is not merely digital but computational.

Courts may encounter:

AI-generated witnesses’ avatars;
synthetic voice communications;
AI-created crime-scene reconstructions;
algorithmically reconstructed events;
digital twins;
biometric simulations;
AI-generated documents;
synthetic surveillance footage; and
automated forensic conclusions.

The courtroom may therefore need to understand not merely whether a file exists, but how an algorithm produced or transformed the information contained in that file.

This raises a profound jurisprudential question:

Can an algorithm become part of the chain of evidence?

Increasingly, the answer will be yes.

34. The Human Element Must Remain Central

Despite technological developments, justice remains a human institution.

AI may assist in:

detecting manipulation;
analysing metadata;
comparing voices;
examining facial consistency;
identifying anomalies;
reconstructing timelines; and
processing enormous quantities of digital information.

But the ultimate determination of:

relevance;
admissibility;
credibility;
reliability;
probative value; and
guilt or liability

must remain within the judicial process established by law.

AI should therefore remain an instrument of forensic assistance, not an autonomous adjudicator.

35. Conclusion

Artificial Intelligence is fundamentally changing the evidentiary landscape.

The law has already moved from paper documents to electronic records. It must now confront the transition from electronic records to synthetic reality.

The greatest challenge presented by deepfakes is not merely technological. It is jurisprudential.

For centuries, courts have developed methods of evaluating human testimony and physical documents. The digital age required courts to develop principles for electronic records. The age of Generative AI now requires a further evolution: principles capable of distinguishing authentic digital evidence from convincingly fabricated digital reality.

The answer should not be technological panic.

Nor should it be blind technological optimism.

The appropriate response lies in a combination of:

statutory compliance, scientific methodology, forensic expertise, provenance verification, chain-of-custody discipline, independent corroboration and judicial scrutiny.

The central principle should remain simple:

The fact that evidence is digital does not make it unreliable; the fact that evidence is visually convincing does not make it authentic; and the fact that evidence is AI-generated does not, by itself, determine its admissibility.

The future of evidence law will therefore depend upon the Court’s ability to distinguish between the existence of a digital file, the authenticity of its provenance, the integrity of its contents, and the truth of the event it purports to depict.

In the age of deepfakes, the judicial question will increasingly change from:

“What does this recording show?”

to:

“How do we know that this recording genuinely represents what it purports to show?”

That question may become one of the defining evidentiary challenges of the twenty-first-century courtroom.

Selected Legal and Technical References

Bharatiya Sakshya Adhiniyam, 2023, particularly Sections 61–63 concerning electronic and digital records.
Anvar P.V. v. P.K. Basheer & Ors., (2014) 10 SCC 473.
Arjun Panditrao Khotkar v. Kailash Kushanrao Gorantyal & Ors., (2020) 7 SCC 1.
Pooja Ramesh Singh v. Jammu and Kashmir Bank Ltd., 2026 INSC 668 — concerning reliance upon AI-generated/hallucinated legal authorities.
National Institute of Standards and Technology (NIST), Reducing Risks Posed by Synthetic Content: An Overview of Technical Approaches to Digital Content Transparency.
NIST, Guardians of Forensic Evidence: Evaluating Analytic Systems Against AI-Generated Deepfakes.
NIST, Examining Digital Media — Synthetic, Repurposed and Deepfake Digital Media.
NIST, Digital Evidence Preservation: Considerations for Evidence Handlers.

Author’s Note

This article is intended as a legal-academic discussion of emerging issues in AI-generated and deepfake evidence. Since the technology and regulatory environment are rapidly evolving, specific statutory provisions, rules, judicial precedents and forensic standards should be verified against the law applicable on the date of publication and the facts of the particular case.