From Witnesses to Algorithms

**Dr. Ishan Atrey

From Eyewitnesses to Electronic Evidence: The Changing Nature of Proof:

Criminal Courts have historically relied heavily on human testimony. An individual witness describes what they “saw,” “heard” or “experienced.” Their testimonial accounts are assessed by the court through examinations and cross-examination. Assessments about the credibility of the witness, about their ability to perceive events, their recollection of those events become integral components of the processes used to establish fact. Although this procedure has not been eliminated, the manner in which criminal evidence is derived is evolving. Nowadays, prosecutors often use video surveillance images, cell phone records, cellular site locations, emails, server logs and many other forms of digital evidence, regardless if anyone personally observed the crime.

In response to these changes in the way to prove the commission of the offence , India has amended its evidence laws. Specifically, the new Bharatiya Sakshya Adhiniyam, 2023 (BSA) has incorporated electronic and digital records into its body of evidence law. Section 61 of the BSA clearly states that electronic or digital records shall not be excluded from admission based solely on their form, whereas section 63 outlines the criteria for admitting electronic records created by computers and communication devices. This is an enormous departure from prior legislation and evidentiary systems that were centred on physical documentation and/or oral testimony. However, recognising electronic evidence as valid does not answer a larger issue: at what point should courts accept what technology records or creates?

The Indian Supreme Court’s jurisprudence under the Indian Evidence Act of 1872 exemplifies why this question arose. Section 65B provided the statutory framework under the Indian Evidence Act, 1872 for proving electronic records produced as computer output. In Anvar P.V. v. P.K. Basheer (2014), the Supreme Court viewed section 65B as the exclusive statutory mechanism for the admissibility of secondary electronic evidence and emphasised that section 65B provided critical protections for the integrity of such materials. Similarly, in Arjun Panditrao Khotkar v. Kailash Kushanrao Gorantyal (2020), the Supreme Court reiterated and applied the same principles outlined in Anvar and analysed the distinctions between primary electronic evidence and computer-generated copies. There was a rational basis for the concern expressed in the jurisprudence regarding electronic evidence: the ease with which electronic evidence could be copied, transferred or modified so as to leave behind physical indicia typically associated with documentary evidence.

However, the experience with section 65B demonstrates an ongoing conflict that continues to exist under the BSA. While provisions designed to limit or prohibit admission of untrustworthy evidence can serve as a safeguard against admitting false or fabricated materials into trials, such provisions can also pose serious obstacles when authenticating electronic evidence intended to be admitted in trials. The problem, however, is not just whether digital evidence can be admitted into court; the question of importance is, does the law provide adequate means for ascertaining that the digital evidence actually proves something and the degree to which reliance may appropriately be put in such evidence.

When the Machine Becomes the Witness: What Can Technology Actually Prove?

When the use of technology extends beyond preserving an event, there will be greater challenges. For example, a CCTV camera could capture what happened in front of it. A telecommunication network could generate a call detail record (CDR), which would document calls made by a telephone user. A mobile phone device could produce location information based on cell towers. In all cases, technology generally operates as a type of recording device. However, newer types of technologies offer the potential to do more. Technologies employing facial recognition can take images and match them against individuals. Other computational technologies can evaluate substantial amounts of data and create patterns or conclusions that investigators can bring before a court. That distinction matters since the further technology proceeds away from recording events to analysing them, the more difficult it is to define specifically what the court is expected to consider as “evidence”.

Although the BSA permits qualified, computer-generated evidence to be introduced into evidence as an alternative to the original evidence, it will not provide any insight into how technology actually works. Although a video recording may show a person entering a building, it does not demonstrate by itself that the person is the Offender. In Chattan Dabas v. State (Nct. Of Delhi) (2025),  where questions were raised regarding the clarity of the Offender’s face in relation to the CCTV recording; the quality of the video recording, and the reliability of a facial recognition report. The court decision demonstrates how identification using technological evidence can include multiple different types of inferential steps. This distinction is important, although the camera recorded the image, the determination of who is identified by the image came from either an investigative staff member, an expert or an algorithm.

A CDR may indicate where a cellular telephone was connected to a network, but it does not necessarily establish that the subscriber personally possessed the device or was physically present at the precise location of the crime. Its evidentiary value therefore depends not only on the authenticity of the record but also on the inference drawn from it.

Additionally, this becomes even more relevant when an algorithm produces a part of an evidentiary conclusion. An algorithm is a collection of computational commands utilised to analyse data and generate an output. If an algorithm generates an output that contributes to establishing the identity of a suspect, reconstructing an event, or associating a person with a crime scene, then the traditional question “who saw this?” no longer applies. Rather, it is transformed into a number of additional inquiries, like what types of data were analysed by the algorithm, how testing was done on the system, what are the limitations of the system, how the output was developed, and whether the defendant can effectively challenge the processes leading to the development of the output.

There have been recent Indian judicial decisions illustrating that these issues are not simply speculative. The courts have begun to encounter controversies relating to CCTVs, facial recognition software, electronic documents and access to supporting materials related to prosecution evidence. In Kishan Agarwal v. State of Rajasthan (2026), the Rajasthan High Court addressed the authenticity of electronic copies, access to underlying electronic material and the use of hash values in considering the defence’s access to prosecution evidence

The Problem of Trust: Authentication, Reliability and the Right to Challenge:

Technological evidence requires courts to assess not only what a system recorded or produced, but also the process through which the result was generated. The central issue is therefore not simply whether technological evidence is authentic, but whether the process that produced it is sufficiently reliable.

Therefore, it is essential to differentiate between the concepts of “authenticity” and “reliability.” Authenticity refers to whether or not a record is what the prosecution asserts it to be. Reliability addresses a different inquiry: assuming that a record is genuine, can the court reasonably depend on that record in drawing the specific conclusions that the prosecution wishes to derive from it? Thus, the probative worth of a record is dependent not solely on the record itself but on the inference that the prosecution attempts to make from it.

It follows that the right of an accused to scrutinise the source documents that comprise the factual basis for the prosecution’s position assumes greater significance in this context. Specifically, in P. Gopalakrishnan v. State of Kerala (2025), the Supreme Court of India addressed an accused’s request for a clone of a memory card containing records that formed the basis for the prosecution’s case. The Court recognised that electronic records included within a prosecution’s case against an accused should be made available to the accused in a way that allows for meaningful defence efforts while also recognising legitimate interests related to protecting privacy and sensitive materials. The Court’s rationale for its decision is significant because it extends the conversation beyond mere admissibility. Simply stated, an accused will be severely handicapped in challenging electronic evidence if the person is denied access to source material necessary to understand/evaluate that evidence.

Similarly, when specialised technologies are utilised as part of an evidentiary process, the difficulty in obtaining relevant source material increases exponentially. For example, if an algorithm is responsible for helping to identify an accused, knowing the outcome of that determination may not be sufficient for a defendant. To engage in meaningful analysis, a defendant may need to know specifics about the data input into the system, specifics about how that data was processed through the algorithm, specifics about error rates associated with previous use of the system, specifics about operating parameters of the system at issue and specifics about how the defendant arrived at a particular result. While this does not mandate disclosure of all proprietary technical details surrounding a system, it raises a broader inquiry: what information must be provided to a defendant so that s/he may effectively challenge results obtained from a technological process?

This produces a possible technological gap between the prosecutor and defence. The investigating agency may have access to the devices from which the data was originated, specialized equipment and software, as well as technical personnel, whereas the defence will likely only receive a copy of the final report. Although the defence technically has the right to dispute the evidence received in this manner, they realistically do not have the ability to adequately test either the underlying data used by investigators to draw conclusions about the defendant’s guilt or the methods by which those conclusions were drawn using technology.

Rethinking Proof: Building Safeguards for Technology-Driven Criminal Trials:

The appropriate response is not to exclude technological evidence from criminal proceedings. It would be unrealistic to exclude digital evidence now, given its deep integration into both the investigation and adjudication processes. More than anything else, the relevant question is whether the current legal structure allows judges and defendants to evaluate the quality of the evidence presented electronically at some point before that evidence determines their fate. This is precisely why the recognition of electronic and digital records in the BSA has been an important step toward this end. In order to further ensure that the admissibility of electronic evidence will be supported by standards designed to protect the integrity, reliability and contestability of digital evidence, there needs to be a focus on providing clear distinctions among types of digital evidence.

Technological evidence differs in the extent to which technology intervenes between an event and the information presented to the court. A recording preserves an event, whereas a technological system may generate or interpret information from processed data. The greater the technological intervention, the greater the need to examine how the information was produced and what assumptions shaped it. Treating all such material simply as ‘electronic evidence’ may therefore overlook the different ways in which errors can arise. The legal framework should therefore distinguish between the statutory admissibility of an electronic record and the extent to which technological processes have shaped the information presented to the court. 

Digital evidence must also be preserved in a way that allows its origin and subsequent handling to be traced. A hash value can help establish whether a digital file has remained unchanged after copying or transmission, but it cannot establish that the original recording accurately captured the incident itself.

The Defendant must also have a reasonable opportunity to scrutinise technological evidence relied upon by prosecutors. The principles recognised in P. Gopalakrishnan v. State of Kerala relating to access to electronic materials required for the preparation of a defence assume even greater importance as investigations become increasingly dependent on technology. Additionally, where applicable, the same principle should apply to any additional information needed to fully comprehend how technologically assisted evidence was developed. In cases where an algorithm played a substantial role in arriving at a determinative evidentiary conclusion, merely producing the ultimate outcome may not be sufficient. The defence may need access to data supporting the development of the algorithm; the methodologies employed; validation procedures; and/or other material required to subject the results to testing.

It does not follow that every line of source code or proprietary technical detail must be divulged in each criminal case. Doing so could raise legitimate concerns regarding privacy rights, cybersecurity, intellectual property, and investigative techniques. A more viable standard is one of meaningful contestability, as Defendants should obtain sufficient information to enable them to recognise and dispute material weaknesses in the technological evidence upon which the prosecution relies.

Furthermore, courts must remain wary of treating technological outputs as being objectively definitive. At times, an impression of precision may conceal fundamental assumptions underlying a technological process. Technological systems and human witnesses can both produce errors. However, the types of errors differ: whereas a witness’s testimony may be subject to error due to perceptual and memory processes, technological output requires that the original input data, as well as the method used for analysis, and the operating conditions in which the system operated, be scrutinized.

Lastly, technological evidence may necessitate a heightened degree of judicial interaction with expert testimony. Judges need not become computer scientists; nonetheless, judges must acquire sufficient knowledge of limitations inherent within technological systems prior to assigning great significance to their outputs. In addition, where there is a dispute surrounding technological processes, independent expertise may prove beneficial for defendants who often have less access to technical capabilities than prosecutors.

The broader principle is straightforward: technology should strengthen the search for truth without becoming a shortcut around the ordinary disciplines of proof. The BSA provides a framework for admitting digital evidence; evidentiary policy must now ensure that courts can also assess its reliability and limits

The future criminal trial may contain fewer moments in which a human witness says, “I saw this happen.” It may contain more moments in which a device, database or algorithm provides an account of what happened. That development is neither inherently beneficial nor inherently harmful. Its legitimacy will depend on whether the legal system retains the ability to question the evidence, examine its origins and expose its weaknesses. The machine may increasingly become part of the evidentiary chain, but the responsibility for deciding what amounts to proof must remain with the court.

**Dr. Ishan Atrey has been actively engaged in legal education and research for over seven years. He currently serves as an Assistant Professor at the School of Law, GD Goenka University, Sohna, Gurugram.

**Disclaimer: The views expressed in this blog do not necessarily align with the views of the Vidhi Centre for Legal Policy.