AI-Generated Evidence: Legal Challenges in India

Artificial intelligence is reshaping the way evidence is gathered, processed, and presented in legal proceedings. Indian courts and investigating agencies now encounter a wide range of AI-generated or AI-processed materials: facial recognition outputs, audio-visual deepfakes, predictive analytics results, AI-assisted document review logs, social media content moderation flags, and algorithmic trading records. These materials raise fundamental evidentiary questions that Indian procedural law even as updated by the Bharatiya Sakshya Adhiniyam, 2023 (BSA) has not yet fully addressed. Counsel engaging with AI-generated evidence must understand where the existing framework applies, where it falls short, and what forensic standards are needed to build or challenge an evidentiary record.

The Statutory Framework for Electronic Evidence

The Bharatiya Sakshya Adhiniyam, 2023, which replaced the Indian Evidence Act, 1872 with effect from 1 July 2024, preserves and expands the framework for admissibility of electronic records. Section 57 of the BSA (corresponding to Section 65B of the Indian Evidence Act) provides that any information contained in an electronic record which is printed on paper, stored, recorded, or copied in optical or magnetic media produced by a computer shall be deemed to be a document and shall be admissible in evidence if the conditions specified in the section are satisfied. Those conditions require, among other things, that the computer was regularly used to store or process information for the purposes of lawful activities of the person with lawful control over it, and that the information was derived in the ordinary course of such activities.

The Supreme Court's decision in Arjun Panditrao Khotkar v. Kailash Kushanrao Gorantyal (2020) 7 SCC 1 settled that a certificate under Section 65B (now Section 57 BSA) from a person in a responsible official position is a mandatory precondition to the admissibility of electronic evidence. This ruling has significant implications for AI-generated outputs, because the person certifying the record must attest that the computer or system producing the output was functioning correctly and that the information was not altered. Where an AI model produces an output, the certifying authority must be able to attest to the integrity and functioning of the AI system at the relevant time a technically demanding standard.

Deepfakes and Synthetic Media: The Authenticity Challenge

Deepfake technology AI-synthesised audio and video content that mimics a real person's appearance or voice poses an acute authenticity challenge in criminal and civil proceedings. Indian law has no dedicated statutory provision governing deepfakes, though the Ministry of Electronics & IT's 2023 advisory to intermediaries under the IT Rules, 2021, specifically lists synthetic or manipulated content as prohibited under Rule 3(1)(b). In litigation, a party seeking to rely on video or audio evidence must now contend with the prospect that opposing counsel will commission a forensic AI analysis to detect manipulation artefacts, lighting inconsistencies, or compression characteristics inconsistent with the claimed recording device.

Courts are increasingly willing to appoint technical experts under Section 45 of the BSA to opine on the authenticity of electronic records and AI-generated content. Counsel should proactively commission forensic authenticity analysis before tendering AI-related evidence and should be prepared to disclose the methodology, training data characteristics, and version of any AI tool used to generate or process the evidence.

Explainability and Bias in AI Outputs

A recurring challenge in AI-generated evidence is the opacity of the underlying model. When a facial recognition system identifies a suspect, or a fraud detection algorithm flags a transaction as suspicious, the output is the product of a model that may itself be a black box its decision weights unknown even to the entity deploying it. In Indian adversarial proceedings, opposing counsel may challenge AI evidence on the grounds that the model is incapable of explaining the basis for its output, that the training data was unrepresentative or biased, or that the system was not validated for the population or context in which it was applied.

These challenges are not hypothetical. Indian regulators including the Securities and Exchange Board of India (SEBI) and the Reserve Bank of India (RBI) have issued guidance emphasising the need for explainability in AI and ML models used in regulated activities. The SEBI circular on algorithmic trading and the RBI's guidance on use of AI in credit underwriting both require that model outputs be interpretable and auditable. Where AI-generated evidence originates from a regulated entity's systems, these regulatory standards can become relevant benchmarks in assessing evidentiary reliability.

Practical Steps for Counsel

Counsel tendering AI-generated evidence should ensure that the Section 57 BSA certificate is obtained from a technically competent person who can attest to the specific AI system's functioning, the version of the model at the relevant time, the integrity of the input data, and the absence of manual post-processing. Where the AI output is critical to the case, an independent technical expert should be retained to prepare a report establishing these foundations before the evidence is tendered.

Counsel challenging AI-generated evidence should seek discovery of the model documentation, training data characteristics, accuracy and precision metrics, and any known failure modes. They should request disclosure of whether the AI system was independently audited and, if so, the audit findings. In appropriate cases, an application under the BSA for court-appointed examination of the AI system may be warranted. The rapidly evolving nature of this area makes early forensic engagement and careful pre-trial preparation essential for both sides.