AI-Generated Evidence Creates Legal Accountability Void Under BSA 2023

The Indian legal system is facing an unprecedented challenge: how to treat evidence generated by artificial intelligence? From facial recognition matches and predictive policing scores to AI-reconstructed crime scenes and cloned voice detection reports, machine-generated outputs are increasingly making their way into courtrooms. Yet, as a recent incisive analysis reveals, the current evidentiary framework under the Bharatiya Sakshya Adhiniyam, 2023 (BSA) is ill-equipped to handle these novel forms of proof. The core dilemma is that AI cannot be sworn in, cannot be cross-examined, and cannot explain its reasoning—leaving a troubling accountability void that neither the BSA nor the earlier Indian Evidence Act, 1872 was designed to fill.

The Uncross-Examinable Witness

Traditional evidence law rests on the assumption that every source of proof can be subjected to adversarial testing. A human witness takes an oath and faces cross-examination; a document is proved by its creator or custodian; an expert explains the methodology behind their opinion under Section 45 of the old Act (now Section 39 of the BSA). AI-generated evidence defies every category. As the analysis notes: "A neural network isn't about to tell you why it marked a face as a match or why an individual got a high score when it called them a high risk of recidivism; it is only going to spit out a statistic made of weights that are even as obscure as they may be to their own creators."

The result is a fundamental mismatch. Courts are being asked to rely on assertions from a source that cannot take an oath, cannot be cross-examined, and often cannot explain its own conclusions in a way a judge can evaluate. This is not merely a procedural inconvenience—it strikes at the heart of the adversarial system's commitment to testing evidence through scrutiny.

A Statute Designed for Passive Records

India's evidentiary framework, both under Section 65B of the old Act and its successor Sections 61 to 63 of the BSA, was built for electronic records that passively store and reproduce data—call logs, emails, CCTV footage. As the Supreme Court of India clarified in Anvar P.V. v. P.K. Basheer and confirmed by the Constitution Bench in Arjun Panditrao Khotkar v. Kailash Kushanrao Gorantyal , the certificate required under these provisions ensures the integrity and custody of the output file. But it does not—and cannot—certify the veracity of the inferential process that produced the content.

The analysis underscores this gap: "Section 63 does not presume that the analytical process used to arrive at an AI's conclusion is sound, that there was sufficient representative training data, or that the error rate for the analytical model has been shared. Certification is used to establish both custody and integrity of output file. It does not (or cannot) guarantee the veracity of the argument that produced the contents of that file." This is the precise evidentiary divide that Indian law must now bridge.

Neither Witness, Nor Document, Nor Expert

The classification struggle is acute. AI output cannot be a witness under Sections 118–134 of the BSA because it lacks human declarant capacity. It is not a straightforward document like a photograph or ledger, because it does not capture reality directly—it constructs it through an inferential model. And when introduced via an expert witness, that expert typically testifies about the machine's general reliability, not the specific output's correctness. As the source observes: "In some cases, the machinery is introduced into evidence via an expert witness, who not only gives the jury information about the machine, but also about its general reliability which the expert is not trying to illustrate for the specific case before him — something that defence counsel seldom presses."

The analysis proposes that AI-generated content may fit a new "hybrid category" of material—admissible in appropriate cases, but subject to limitations: it should be treated as corroborative rather than substantive evidence, and only when its working is sufficiently transparent to allow meaningful challenge.

A Four-Point Accountability Matrix

When the machine cannot answer for itself, responsibility must be distributed among the human and institutional actors behind it. The analysis identifies four key areas of accountability that require separate judicial scrutiny:

First, design-time failures —the model developer may have used biased training data, failed to provide confidence bounds, or designed an opaque algorithm. If a forensic tool cannot be validated for the specific population and context in which it is used, no post-hoc disclaimer should absolve its creator.

Second, agency deployment —the police or forensic lab that operates the AI decides whether to use it within validated parameters and whether to treat its output as a mere investigative lead or as definitive proof. Predictive policing scores and facial recognition matches are tools, not findings of fact; misuse of such tools is a problem of implementation, not technology.

Third, the certifying officer's burden under Section 63 of the BSA is distinct. Signing a certificate that an AI-generated record is "as described" without revealing its inferential nature implicitly misrepresents the evidence. The officer must disclose that the output comes from an inferential model, not a passive recording device.

Fourth, the court's gatekeeping role is paramount. Indian courts, the analysis argues, should borrow from the Daubert standard applied in the United States, which requires judges to evaluate the scientific soundness of an expert's method before admitting evidence. Specifically, courts should demand that AI systems disclose known error rates, training data composition, and validation studies conducted in the Indian context before attributing any evidentiary weight. "Failing to explore this output after acceptance by a judge without explanation is not neutrality. It's shutting your eyes to the unaccountable system you've outsourced a judicial determination to."

Lessons from Abroad

Other jurisdictions are beginning to respond. The European Union's AI Act classifies forensic and law-enforcement AI systems as "high-risk," requiring documentation, human oversight, and accuracy testing. Some overseas courts have introduced mandatory disclosure when generative AI is used to create submissions or evidence. India's IT Rules, 2026, take a partial step by requiring traceability of AI-generated content deployed publicly, but they do not address evidentiary value in court.

The analysis observes: "At the moment, there is no such thing in India as compulsory test of forensic AI in litigation." This stands in sharp contrast to the growing prevalence of AI tools in the justice system.

A Path Forward: Disclosure, Corroboration, and Gatekeeping

The author argues that a coherent approach does not require waiting for new legislation—much can be achieved through purposive interpretation of existing provisions and development of procedural practices. Three concrete steps are proposed:

  1. Technical disclosure attached to the certificate – For any AI-generated evidence, the Section 63 certificate should include the AI's validated parameters, documented error rate, and confirmation that it was applied within those parameters.

  2. Corroboration requirement – AI content should ordinarily be admitted as corroborative evidence, not substantive evidence, necessitating independent human verification before it can sustain a conviction or a substantive finding in civil proceedings.

  3. Structured reliability hearing – Courts should establish, through practice directions or eventual amendment, a threshold inquiry akin to a "reliability hearing" before allowing case-determinative AI evidence. The burden of establishing reliability would fall on the party seeking its admission.

Conclusion

The question "who is the witness when artificial intelligence becomes a witness?" does not yield a simple answer. Responsibility must be spread across the developer who created the model, the agency that deployed it, the officer who certified it, and the court that weighed it. But that spread-out responsibility cannot become no responsibility at all because the immediate source is a machine. As the analysis concludes: "The Bharatiya Sakshya Adhiniyam, 2023 did make available to India a modern system of electronic records and records that never imagined that some of them will think for themselves – albeit in a rather coarse way. It is the need of the hour to fill the vacuum created by those standards (disclosure, corroboration, and effective judicial gatekeeping), in Indian evidence law, before it begins the next iterative cycle."

For legal professionals, the message is clear: the era of passively accepting AI output as just another electronic record is over. Proactive engagement with the evidentiary challenges of artificial intelligence is no longer optional—it is a professional imperative.