Author: Avilasha Kumari Jena, Asian School of Business Management, ASBM University
Abstract
The increasing use of generative AI has created a serious evidentiary challenge for Indian courts. AI can now generate realistic videos, clone voices, produce images, and create written reports that appear authoritative but may be false, biased, or manipulated. The Bharatiya Sakshya Adhiniyam, 2023 recognizes electronic records and provides a statutory route for their admissibility through Sections 61, 62, and 63. Section 63, the successor to Section 65B of the Indian Evidence Act, 1872, requires certification and technical particulars for secondary electronic evidence.
However, certification alone does not establish truth. AI-generated evidence must be tested for provenance, integrity, bias, manipulation, and independent corroboration. An algorithm cannot be treated as a witness because it cannot be cross-examined; the human operator or forensic expert must instead explain the technology and its limitations. Judicial precedents such as Anvar P.V. v. P.K. Basheer and Arjun Panditrao Khotkar v. Kailash Kushanrao Gorantyal establish that certification is a mandatory condition for admitting secondary electronic evidence.
This article argues that AI-generated evidence should neither be automatically rejected nor blindly accepted. Courts should treat unverified AI material as corroborative rather than substantive proof until its origin, method of generation, and reliability are transparently established. AI-specific safeguards disclosure of the model and inputs, preservation of metadata, hash verification, independent forensic examination, and cross-examination of human witnesses are necessary to preserve fair trial rights under Articles 14, 20(3), and 21 of the Constitution.
To the Point
Artificial Intelligence (AI) has entered Indian courtrooms not as a litigant or witness, but through the evidence itself synthetic audio, deepfake videos, AI-generated images, algorithmic forensic reports, chatbot transcripts, and facial-recognition outputs. The central legal issue is whether such material can be admitted and relied upon under the Bharatiya SakshyaAdhiniyam, 2023 (BSA).
The answer is that AI-generated evidence is not per se inadmissible. It may be received as an electronic record, provided the party satisfies the procedural requirements of Section 63 BSA. However, admissibility is not equivalent to reliability. A valid certificate, hash value, or expert endorsement proves procedural compliance; it does not conclusively establish that the underlying content is authentic, untampered, unbiased, or truthful.
Use of Legal Jargon
Under Section 61 BSA, an electronic record cannot be denied admissibility merely because it is in electronic form. Section 62 BSA provides that the contents of electronic records may be proved in accordance with Section 63 BSA. Section 63 BSA, which replaces Section 65B of the Indian Evidence Act, 1872, governs the admissibility of electronic records, particularly secondary electronic evidence.
For AI-generated material such as a deepfake video, cloned voice recording, AI transcript, or algorithmic report the producing party must establish:
The provenance of the electronic record;
The device, software, model, or system through which it was generated or extracted;
The integrity of the file through metadata and hash values;
Compliance with the certificate requirement under Section 63(4) BSA;
Independent forensic verification of authenticity and reliability.
An algorithm cannot be examined as a witness because it lacks testimonial competence. It cannot perceive events, take an oath, explain its reasoning, or face cross-examination. Therefore, the evidentiary link must be a competent human witness such as a forensic expert, developer, system administrator, or investigating officer who can depose on the AI tool, input data, output, limitations, and possibility of manipulation.
The constitutional dimension is equally important. Uncritical reliance on unverified AI material may violate Article 21, which guarantees a fair trial; Article 14, where algorithmic bias causes discriminatory treatment; and Article 20(3), where AI-based inference is used to attribute guilt without adequate human corroboration.
The Proof
The statutory framework begins with Sections 61 to 63 BSA. Section 61 removes the presumption that electronic records are inherently weaker than physical documents. Section 63 requires that a secondary electronic record be accompanied by a certificate identifying the record, describing its manner of production, giving particulars of the device, and addressing the statutory conditions for reliability.
Section 63(4) BSA requires certification by a person responsible for the device or the relevant activity. The certificate must identify the electronic record, explain how it was produced, disclose particulars of the device, and provide relevant technical particulars, including hash value and hash algorithm where applicable.
This framework is directly relevant to AI evidence. A screenshot of an AI-generated video, a forwarded WhatsApp file, or a pen drive containing synthetic audio is not self-authenticating. The court must examine:
Which AI model or tool generated the material;
The version of the software or algorithm;
The source and quality of input data;
Whether the voice, face, or image was cloned or synthesized;
Whether metadata, device logs, cloud records, compression patterns, watermarks, and hash values support authenticity;
Whether an independent forensic expert can verify or rebut the material.
For instance, if the prosecution produces an AI-generated voice recording alleged to contain a confession, a Section 63 certificate may make the recording admissible, but it cannot by itself prove that the accused actually made the statement. The court must test the origin of the audio, the possibility of voice cloning, metadata consistency, chain of custody, and corroborative evidence.
Deepfakes intensify this problem. A deepfake may fabricate a confession, falsely place an accused at a crime scene, or create a misleading identification. Conversely, a genuine video may be falsely branded as a deepfake to manufacture reasonable doubt. Courts must therefore not decide authenticity solely by visual or auditory impression; forensic and technical scrutiny is indispensable.
Case Laws
Anvar P.V. v. P.K. Basheer, (2014) 10 SCC 473
The Supreme Court held that Section 65B of the Indian Evidence Act, 1872 is a complete code governing the admissibility of secondary electronic evidence. A certificate under Section 65B (4) is mandatory, and oral evidence cannot substitute for it. The Court held that electronic records produced without the required certificate are inadmissible.
This principle continues to guide the interpretation of Section 63 BSA. In the AI context, it means that an AI-generated video, cloned voice recording, or algorithmic report cannot be admitted merely because it exists in digital form; statutory certification and provenance must be established.
Arjun Panditrao Khotkar v. Kailash Kushanrao Gorantyal, (2020) 7 SCC 1
A three-judge Bench of the Supreme Court reaffirmed that a Section 65B (4) certificate is a condition precedent for the admissibility of secondary electronic evidence. The Court clarified that the certificate cannot be replaced by oral evidence, though limited exceptions may arise where the original record is produced or where the party demonstrates due diligence in attempting to obtain the certificate.
This decision strengthens the proposition that AI-related electronic evidence must satisfy strict procedural safeguards before it is received by a court.
Judicial Approach under Section 63 BSA
The transition from Section 65B of the Evidence Act to Section 63 BSA does not dilute the need for authentication. Rather, it reinforces the requirement of identifying the electronic record, disclosing the manner of production, providing device particulars, and producing the prescribed certificate.
For AI evidence, this means courts must move beyond mere certification and adopt scientific authentication. Hash values may show that a file has not been altered after a certain point, but they cannot prove that the original file was genuine before that point. If the source material itself was synthetic, a valid hash value alone cannot establish reliability.
Conclusion
AI-generated evidence can be placed before Indian courts, but it cannot be trusted merely because it is digital, technical, or visually convincing. The BSA permits electronic records to be admitted, and Section 63 provides the procedural mechanism for proving them. Judicial precedents also insist on certification and strict compliance with electronic-evidence requirements.
Yet admissibility is only the first stage. Courts must separately examine authenticity, provenance, reliability, bias, manipulation, and corroboration. An algorithm may assist in generating or analyzing material, but it cannot replace human testimony, forensic verification, or judicial reasoning. The accused’s right to a fair trial requires that AI-related material be open to challenge through examination of the human expert or operator responsible for it.
The governing principle should be clear: AI-generated evidence is admissible only when its origin, method of creation, and reliability are transparent and independently verifiable. Until India enacts clearer AI-specific evidentiary rules, courts should treat unverified AI material as corroborative evidence, not as automatic proof of truth.
FAQs
Q1. Is AI-generated evidence admissible in Indian courts?
Yes. AI-generated text, audio, video, images, transcripts, and forensic reports may be admitted as electronic records under Section 63 BSA, provided the statutory requirements are satisfied.
Q2. Does a Section 63 BSA certificate prove that AI evidence is true?
No. The certificate establishes procedural compliance for admissibility. It does not itself establish accuracy, authenticity, or probative value.
Q3. Can an AI algorithm be cross-examined?
No. An algorithm lacks legal personality and testimonial competence. The human operator, developer, or forensic expert must instead be examined and cross-examined.
Q4. Are deepfakes admissible in court?
A deepfake may be technically admissible as an electronic record, but it must undergo forensic authentication. If it is found to be manipulated or synthetic, it may be rejected or given no evidentiary value.
Q5. What safeguards should courts insist upon for AI evidence?
Courts should require disclosure of the AI model and version, source data, metadata, hash values, device logs, chain of custody, independent forensic examination, and cross-examination of the human witness responsible for producing or verifying the material.
Q6. Can AI evidence alone justify a conviction?
Ordinarily, no. Until its authenticity and reliability are independently established, unverified AI material should be treated as corroborative rather than substantive evidence.
