Author: MD Nasiur Rahaman Khan
College: St. Joseph’s College of Law, Bengaluru, Karnataka
LinkedIn Link: https://www.linkedin.com/in/md-nasiur-rahaman-khan-3930b5264
1. Abstract:
Artificial Intelligence (AI) is no longer isolated from the legal world, the emergence of AI has simplified all our complex legal tasks; it has quietly entered through the legal system through platforms such as ChatGPT, Gemini, Perplexity and Claude. As courts increasingly are relyingupon electronically generated information, a difficult evidentiary question emerges today: can AI-generated outputs qualify as expert evidence under the Bharatiya Sakshya Adhiniyam (BSA), 2023? While the new evidence regime modernizes the admissibility of electronic records through Sections 39 and 63, it still remains rooted in human accountability, procedural fairness, and constitutional due process. My article critically examines whether AI systems can satisfy the legal standards governing expert opinion and electronic evidence in India. It contends that even if AI cannot act as an expert witness on its own, its outputs could be acceptable if they are verified as electronic documents and evaluated by trained human experts. My article contends that justice may use algorithms, but it cannot abandon human responsibility fully. After all, a courtroom may tolerate a clever algorithm, but one thing that nobody wants is: “Justice outsourced to a machine.”
2. To the Point:
“Justice is the perpetual will to give each his due; let its pursuit be open, corroborated, and heard.”
– Ulpian, as cited in Justinian’s Institutes, I.1.1
The Maxim is old but today it takes a new importance in an age where proof may originate not from a human witness, but from an algorithmic process. Modern justice demands more than computational accuracy; it requires provenance of data, corroboration of conclusions, explainability of reasoning, and above all, the right of the parties to contest algorithmic evidence through meaningful and fair scrutiny. A system that cannot be questioned, a system that cannot be examined risks replacing adjudication with automation. The rise of AI has led to what I describe as the ‘algorithmic witness,’compelling courts to reconsider whether technological outputs can satisfy the foundational guarantees of fairness, transparency, and accountability which are embedded within the evidence law.
Today AI shapes justice through surveillance and software, but its reasoning is often opaque. Indian evidence law still insists on human expertise, human accountability, and the human intellect of cross-examination. An algorithm can produce an answer, but it cannot raise its hand under oath. This distinction lies at the heart of the debate.
3. Use of Legal Jargon
3.1 Statutory framework
The BSA, 2023 modernizes Indian evidence law while preserving safeguards. Section 39(1) makes expert opinion significant in law, science, art, handwriting, and fingerprints; Section 39(2) extends this to the Examiner of Electronic Evidence under IT Act §79A. Section 61 affirms electronic records cannot be excluded merely for being electronic, and Section 63 requires compliance with admissibility conditions. The IT Act empowers the government to appoint certified examiners, ensuring digital evidence is admitted only through human oversight.
3.2 Core legal proposition
AI cannot be a “person” in the legal sense, and therefore cannot testify like a human witness. The witness provisions of the BSA are framed around persons who can actually understand questions and give rational answers; even a witness unable to speak is still treated as a human witness who uses signs, writing, or gestures to convey information. AI may generate useful output or suggestions, but in law, useful does not mean admissibility.
So, the real question is not whether AI can speak in the court. It is whether its output can survive in the courtroom’s old but unforgiving test: examination,responsibility, obligation, authenticity, reliability, and accountability, if it is satisfying these criteria then the law may permit its assistance.
4. The Proof
4.1 The black box problem
Not all algorithms deserve the same courtroom courtesy. The real danger is opacity: an AI system may deliver a clean solid conclusion while hiding the path it took to get there. In evidence law, that is a serious problem, because courts don’t want merely an answer; they want a reasoned, testable, and challengeable answer. The BSA, 2023 still keeps expert opinion tied to human competency, given that transparency is not a decorative virtue here; it is the price of admissibility.
4.2 The three types of AI
A. Generative AI such as ChatGPT, Claude, and Gemini is the weakest evidentiary category. It can produce polished language, but cannot be proof. If a system can generate fluent output without a human guarantee of accuracy, its value in court becomes secondary or in worst scenario misleading.
B. Forensic AI tools such as facial recognition, CCTV analytics, DNA software, or cyber-forensic platforms stand on better ground. These tools are more scientifically grounded and can support investigation or expert assessment, but even here the law insists on a human intermediary. Section 39(2) expressly treats the Examiner of Electronic Evidence as an expert, which fits the model of AI-assisted analysis validated by thequalified person rather than by the machine alone.
C. Predictive AI used for bail prediction, recidivism scoring, or risk profiling is most constitutionally dangerous. Once a machine starts suggesting who may offend, the presumption of innocence becomes dangerous. That is why Article 21 matters so much: fairness, dignity, and substantive due process cannot be replaced by a confidence score with an AI interface. The constitutional anxiety cannot be imaginary.
5. Case Laws
5.1 Anvar P.V. v. P.K. Basheer
This landmark judgment transformed the law relating to electronic evidence by holding that electronic records are admissible only when the statutory requirements governing certification are satisfied. The decision emphasized on the authenticity and reliability, principles that become even more significant when AI-generated outputs are sought to be introduced before a court.
5.2 Arjun Panditrao Khotkar v. Kailash Kushanrao Gorantyal
The Supreme Court reaffirmed the rule laid down in Anvar P.V. and clarified that compliance with certification requirements is mandatory. The case demonstrates that technological evidence cannot bypass procedural safeguards merely because it appears sophisticated or reliable
5.3 Ramesh Chandra Agrawal v. Regency Hospital Ltd.
The Court observed that expert evidence is advisory in nature and derives its value from the competence, qualifications, and methodology of the expert. This principle creates a significant obstacle for AI systems, which may generate conclusions but cannot independently establish expertise or accountability.
5.4 State of Himachal Pradesh v. Jai Lal & Ors.
The Supreme Court held that expert opinion must be founded upon recognized principles and cannot be based on speculation alone. The decision is particularly relevant to AI-generated outputs, as courts must be satisfied thatthe underlying methodology is reliable and capable of scrutiny.
5.5 Selvi v. State of Karnataka
While dealing with narco-analysis, polygraph tests, and other scientific techniques, the Court emphasized that technological or scientific methods cannot override constitutional protections. The judgment reinforces the importance of Article 21 when evaluating emerging forms of evidence.
6. Conclusion
The conclusion is simple, even if the technology is not. AI is not itself an expert witness under the BSA because it is not a “person” competent to testify. Its output can assist a human expert, but its admissibility will still depend on authenticity, reliability, and compliance with the statutory safeguards governing electronic records. That is the law’s compromise: use the machine, but do not surrender the judgment.
Therefore, AI cannot independently stand in the witness box as an expert. In my view, at best, it may serve as a powerful assistant whose outputs are authenticated as electronic records and validated by a qualified human expert. As India moves deeper into the age of algorithms, legislative and judicial safeguards will become indispensable.
Justice has always demanded more than answers, it demands reasons, it demands fairness. Until an algorithm can explain itself and be held accountable for its mistakes, the final witness in every courtroom will remain a human.
7. FAQs
Q1. Can Artificial Intelligence testify in Indian courts?
No. Under the BSA, 2023, testimony is given by persons who are legally competent to testify. AI systems do not possess legal personality and therefore cannot independently testify.
Q2. Can AI-generated outputs be admitted as evidence?
Potentially yes. AI-generated outputs may be admitted as electronic evidence if they satisfy the requirements of authenticity, reliability, and certification under the applicable provisions of the BSA, 2023.
Q3. What is the significance of Section 39 of the BSA, 2023?
Section 39 makes expert opinions relevant facts in judicial proceedings. It enables courts to take into account the opinions of highly qualified individuals, such as specialists in developing fields of knowledge and electronic evidence.
Q4. Why is the “black box problem” important in AI evidence?
The black box problem refers to the difficulty of understanding how certain AI systems work in the background and reach their conclusions. If courts cannot scrutinize the reasoning behind an AI-generated output, then concerns arise regarding transparency, fairness, and the right to a fair trial under Article 21 of the Constitution.
8. References
Primary Sources – Legislation
Bharatiya Sakshya Adhiniyam 2023 (India). New Delhi: Ministry of Law and Justice.
Information Technology Act 2000 (India). New Delhi: Ministry of Law and Justice.
Constitution of India 1950. New Delhi: Government of India.
Primary Sources – Cases
Anvar P.V. v P.K. Basheer (2014) 10 SCC 473 (Supreme Court of India).
Arjun Panditrao Khotkar v Kailash Kushanrao Gorantyal (2020) 7 SCC 1 (Supreme Court of India).
Ramesh Chandra Agrawal v Regency Hospital Ltd and Others (2009) 9 SCC 709 (Supreme Court of India).
State of Himachal Pradesh v Jai Lal and Others (1999) 7 SCC 280 (Supreme Court of India).
Selvi v State of Karnataka (2010) 7 SCC 263 (Supreme Court of India).
Secondary Sources – Academic Articles
Parihar, K. (2025) Admissibility of AI Generated Forensic Evidence Under Section 39 of the BSA. SSRN Working Paper (21 October 2025).
Shekar, J. (2024) The Admissibility of Scientific Evidence in the Bharatiya Sakshya Adhiniyam, 2023 and the Need for an Indian Daubert. SSRN Working Paper (11 August 2024).
Bharati, R.K. and Nagarale, S. (2024) Digital Forensic Science and Evidentiary Standards in the Bharatiya Sakshya Adhiniyam (BSA) 2023: A Legal Examination of Admissibility. SSRN Working Paper (21 November 2024).
Secondary Sources – Online Legal Commentary
LiveLaw (2026) ‘AI in digital forensics: are Indian evidence laws equipped to handle machine-generated proof?’, LiveLaw, 27 April.
Available at: https://www.livelaw.in/articles/digital-forensics-indian-evidence-laws-ai-generated-proof-531826 (Accessed: 4 June 2026).
Bar and Bench (2026) ‘AI in Indian litigation: the quest for precision and verifiability’, Bar and Bench, 7 April.
Available at: https://www.barandbench.com/view-point/ai-in-indian-litigation-the-quest-for-precision-and-verifiability (Accessed: 5 June 2026).
