Introduction
Artificial Intelligence has moved from the realm of science fiction into the everyday machinery of justice. AI-generated evidence in Indian courts raises difficult questions about authenticity, reliability and accountability. When artificial intelligence can interpret, reconstruct and generate information, who should bear responsibility when its output is wrong? It can analyse thousands of documents, identify patterns in images, recognise voices, enhance photographs, detect alterations and even reconstruct events from fragmented digital material. Yet, as artificial intelligence increasingly finds its way into investigations and litigation, an uncomfortable question emerges: when an AI-generated output is presented as evidence, who assumes responsibility for its accuracy?
The machine?
The developer?
The investigator?
The forensic expert?
Or the lawyer who places it before the court?
The answer is neither technologically simple nor legally settled.
Indian courts have already developed a substantial jurisprudence concerning electronic evidence. With the coming into force of the Bharatiya Sakshya Adhiniyam, 2023, the legal framework governing electronic and digital records has acquired renewed statutory significance. However, AI-generated material is fundamentally more complex than ordinary electronic records because artificial intelligence does not merely preserve information—it can interpret, reconstruct, predict, and generate information.
The courtroom must therefore confront a new evidentiary dilemma: should an algorithmic conclusion be trusted merely because it appears precise?
The answer must be an emphatic no.
From Digital Evidence to AI-Generated Evidence
Electronic evidence is not new to Indian courts.
Emails, CCTV recordings, mobile-phone data, computer files, server logs and digital photographs have long been considered within the evidentiary framework. The Bharatiya Sakshya Adhiniyam, 2023 expressly recognises electronic and digital records and provides statutory requirements concerning their proof. Sections 61, 62 and 63 are particularly significant in this regard.
But AI changes the nature of the problem.
Consider a CCTV recording showing a person entering a building. The recording is a digital record of an event.
Now imagine an AI system examining that recording and concluding that the person is the accused.
The CCTV footage records.
The AI interprets.
That distinction is crucial.
The reliability of the original recording and the reliability of the AI’s interpretation are two separate evidentiary questions. A court must not allow the apparent sophistication of an algorithm to collapse those questions into one.
Can AI-Generated Material Be Admitted as Evidence?
The existence of an AI component does not automatically render digital material inadmissible.
The BSA recognises electronic and digital records as capable of being proved in accordance with the statutory requirements. Section 63 lays down conditions concerning the manner in which electronic records are produced and authenticated.
However, admissibility is not the same as evidentiary weight.
A court may admit material and subsequently determine how much reliance should be placed upon it.
This distinction becomes especially important where artificial intelligence has modified or interpreted the underlying material.
An AI-enhanced photograph, for example, may improve visibility. But if the enhancement creates details that were not present in the original image, the resulting picture cannot automatically be treated as an accurate representation of reality.
Similarly, an AI-generated transcription may be useful investigative material, but its accuracy cannot simply be presumed because it was produced by sophisticated software.
The central judicial question should remain:
What exactly did the AI do to the evidence?
The Supreme Court’s Emerging Approach to AI
The Supreme Court has recently provided an important indication of how Indian courts may approach AI-generated material.
In Pooja Ramesh Singh v. Jammu and Kashmir Bank Ltd., 2026 INSC 668, the Supreme Court dealt with the problem of AI-generated hallucinations appearing in judicial material. The Court expressed serious concern about fabricated or non-existent authorities being generated through AI and relied upon without verification.
The significance of the judgment extends beyond legal research.
Its underlying message is powerful:
AI output is not self-authenticating.
A machine can generate an answer that looks authoritative while being completely wrong.
The Court’s approach demonstrates that technological convenience cannot displace the judicial duty of verification.
If an AI system can invent a case citation, it can also potentially misidentify a face, incorrectly interpret an audio recording, or generate a misleading reconstruction of an event.
The lesson is therefore universal: the more consequential the AI output, the more rigorous the human verification must be.
The Human Being Behind the Algorithm
The greatest difficulty with AI evidence is that artificial intelligence cannot itself assume legal responsibility.
It cannot enter the witness box.
It cannot take an oath.
It cannot be cross-examined.
It cannot explain why it produced a particular result.
And it certainly cannot be punished for misleading a court.
Responsibility must therefore remain with the human beings and institutions that deploy and rely upon the technology.
Suppose an AI-based facial-recognition system identifies an accused person with a purportedly high degree of confidence. If that identification later proves incorrect, the mere statement that “the software identified him” cannot end the inquiry.
The court should be able to ask:
Who operated the system?
What software was used?
What data was supplied?
What methodology was followed?
What is the known error rate?
Was the system independently tested?
Was the original material preserved?
Was the result reviewed by a qualified expert?
These are not questions designed to obstruct technological progress.
They are questions designed to protect judicial reliability.
The Black-Box Problem
Artificial intelligence creates another formidable challenge: the black-box problem.
Some AI systems can produce highly sophisticated conclusions without providing an explanation that a human being can easily understand.
This creates a fundamental difficulty in the courtroom.
Law demands reasons.
Evidence demands scrutiny.
Cross-examination demands an opportunity to challenge the basis of an assertion.
But what happens when the system itself cannot adequately explain how it reached its conclusion?
A courtroom cannot simply accept:
“The algorithm says so.”
That would transform technological output into an evidentiary oracle.
The law has never treated evidence as credible merely because it is complicated. If anything, complexity demands greater scrutiny.
Anvar P.V. and the Integrity of Electronic Evidence
The Supreme Court’s decision in Anvar P.V. v. P.K. Basheer, (2014) 10 SCC 473, remains a cornerstone of Indian electronic-evidence jurisprudence.
The Court recognised the special concerns surrounding electronic records and insisted upon statutory safeguards for their proof.
The Constitution Bench decision in Arjun Panditrao Khotkar v. Kailash Kushanrao Gorantyal, (2020) 7 SCC 1, subsequently clarified important aspects concerning electronic evidence and the certificate mechanism under the former Section 65B of the Indian Evidence Act.
Although neither case concerned generative AI in its present form, their underlying principle remains highly relevant.
Digital evidence must possess demonstrable authenticity and reliability before it can become a dependable foundation for judicial findings.
AI does not diminish that requirement.
It strengthens it.
AI Evidence and the Burden of Proof
The emergence of AI also raises a constitutional concern.
Imagine that an AI system declares that a particular person is responsible for an offence with a “99% confidence score”.
The number sounds impressive.
But what does it actually mean?
Does it represent the probability that the identification is correct?
Does it reflect the accuracy of the underlying database?
Does it account for lighting, camera quality, facial angle, or demographic variation?
A numerical confidence score without methodological context may create an illusion of scientific certainty.
Indian criminal jurisprudence is built upon the principle that the prosecution must establish guilt according to the applicable standard of proof. AI should not quietly reverse that burden by compelling an accused person to disprove an algorithmic conclusion.
The accused should not have to prove that the machine is wrong merely because the machine sounds confident.
Deepfakes: The New Evidentiary Battlefield
The deepfake phenomenon may become one of the greatest challenges for courts.
A manipulated video can depict a person saying words they never uttered.
A cloned voice can imitate a person’s speech.
An AI-generated photograph can place an individual in a location where they never existed.
The problem is particularly acute because visual evidence traditionally carries enormous persuasive force.
A judge, like any human being, may instinctively regard a photograph or video as compelling.
But in the age of generative AI, seeing is no longer necessarily believing.
A digital video must therefore be examined for provenance, metadata, continuity, alteration, and authenticity where its genuineness is disputed.
AI itself may assist in detecting deepfakes, but this produces an intriguing evidentiary circle:
AI generates the deception, and AI may be asked to detect it.
The human expert and judicial scrutiny therefore remain indispensable.
Who Should Bear Responsibility?
A future Indian framework for AI-generated evidence should clearly distinguish different forms of responsibility.
The investigating agency should remain responsible for ensuring that AI-generated investigative material is not treated as conclusive without proper verification.
The forensic expert should be responsible for accurately explaining the methodology, limitations and reliability of the AI tool used.
The organisation deploying the system should maintain adequate safeguards, documentation and audit mechanisms.
The technology provider may bear responsibility where the law establishes that defective design, misleading representations or negligent conduct caused legally cognisable harm.
Finally, the legal professional presenting AI-generated material cannot outsource professional responsibility to an algorithm. Material placed before a court must be appropriately verified.
The common thread is simple:
AI may assist in producing evidence, but responsibility cannot be delegated to a machine.
AI Should Assist the Court, Not Replace the Court
There is a temptation to regard AI as a solution to the problem of human error.
That temptation must be resisted.
Human beings make mistakes.
So do algorithms.
The difference is that an algorithmic mistake can acquire an extraordinary aura of objectivity because it is expressed through numbers, technical language and computational precision.
The judicial process must therefore preserve the human capacity to question technological conclusions.
AI can assist a judge.
It can help identify patterns.
It can analyse enormous quantities of material.
It can flag inconsistencies.
But the final determination of credibility, guilt and liability must remain an exercise of judicial reasoning.
The Need for an AI Evidence Protocol in India
The increasing use of AI suggests that India may eventually require specific procedural guidelines governing AI-generated evidence.
Such a framework could require disclosure of the AI system used, the original data supplied, the methodology adopted, the extent of human intervention, known limitations, and relevant accuracy statistics.
Parties relying upon AI-generated evidence could also be required to preserve the original material so that the court can compare the source with the AI-generated output.
Independent expert verification may be particularly important where AI evidence is central to the prosecution or defence case.
Such safeguards would not represent hostility towards innovation.
Quite the contrary.
The best way to make technological evidence acceptable is to make its reliability demonstrable.
The Road Ahead
The Indian judiciary stands at the threshold of a new evidentiary era.
The challenge is not whether courts should reject artificial intelligence.
That would be neither realistic nor desirable.
The real challenge is determining how much trust the law should place in machine-generated conclusions and under what conditions.
The answer must be guided by familiar principles: authenticity, reliability, transparency, fairness and the opportunity to challenge evidence.
AI should not receive a presumption of infallibility simply because it operates at computational speed.
Nor should it be rejected merely because it is technologically novel.
The courtroom must occupy the middle ground—open to innovation, but sceptical of unverified certainty.
Conclusion
Artificial intelligence has the capacity to transform the way evidence is collected, analysed and presented before Indian courts.
But it also has the capacity to manufacture convincing falsehoods at a scale and speed previously unimaginable.
That makes accountability the central question.
Who bears responsibility when AI-generated evidence is wrong?
The answer should never be: the machine.
The responsibility must ultimately rest upon the human beings and institutions that choose to create, deploy, verify, and present that material.
The Supreme Court’s approach to unverified AI-generated material offers a valuable judicial warning: technological sophistication cannot substitute for truth, and computational confidence cannot replace proof.
As Indian evidence law moves further into the age of algorithms, one principle deserves to remain immutable:
“Artificial intelligence may generate the evidence; only accountable human judgment can give it legal meaning.”
The future courtroom may be assisted by algorithms, but it must never become subordinate to them.
Because justice is not merely about discovering what a machine says happened.
It is about establishing, through reliable and challengeable evidence, what actually happened.
Landmark Judgments
Anvar P.V. v. P.K. Basheer, (2014) 10 SCC 473
A foundational Supreme Court decision concerning the admissibility and authentication of electronic evidence.
Arjun Panditrao Khotkar v. Kailash Kushanrao Gorantyal, (2020) 7 SCC 1
A Constitution Bench judgment clarifying the requirements governing electronic records under the earlier Evidence Act and the role of the statutory certificate.
Tomaso Bruno v. State of Uttar Pradesh, (2015) 7 SCC 178
The Supreme Court recognised the growing importance of scientific and electronic evidence in the administration of criminal justice.
Pooja Ramesh Singh v. Jammu and Kashmir Bank Ltd., 2026 INSC 668
A significant recent decision highlighting the dangers of relying upon unverified AI-generated material and hallucinated authorities.
References
Bharatiya Sakshya Adhiniyam, 2023, Sections 39, 61, 62 and 63.
Anvar P.V. v. P.K. Basheer, (2014) 10 SCC 473.
Arjun Panditrao Khotkar v. Kailash Kushanrao Gorantyal, (2020) 7 SCC 1.
Tomaso Bruno v. State of Uttar Pradesh, (2015) 7 SCC 178.
Pooja Ramesh Singh v. Jammu and Kashmir Bank Ltd., 2026 INSC 668.
Supreme Court of India, Landmark Judgment Summaries.
Ministry of Law and Justice, Government of India, Bharatiya Sakshya Adhiniyam, 2023.
