Can an Accused Demand the Prosecution’s AI Tools?
By Rajnandini Verma
Abstract
Artificial intelligence is changing the way information can be examined during a criminal investigation. A system can sort large amounts of digital material, identify patterns or assist investigators in locating relevant information within a huge dataset. But when such technology becomes part of an investigation, a difficult question follows: how much of that technological process should be visible to the accused?
Indian criminal procedure does not currently recognise a separate right allowing every accused person to demand the source code, model or internal workings of an AI system used by investigators. At the same time, technological secrecy cannot be used to defeat the accused’s existing rights to relevant material and a fair trial. The legal issue is therefore less about ownership of an algorithm and more about whether the algorithm has affected evidence that the accused must be able to challenge.
Introduction
Criminal trials are based on evidence, not on the reputation of the technology used to find that evidence.
That principle becomes important when artificial intelligence enters an investigation.
Imagine that investigators have thousands of messages, documents or digital records. An AI system is used to identify particular patterns and the investigators subsequently rely on those results. The accused then asks a simple question: “What exactly did the system find, and what information was it working with?”
There is no provision in the Bharatiya Nagarik Suraksha Sanhita, 2023 (BNSS) that says an accused has an automatic right to inspect every AI system used by the police.
But there is also no rule saying that evidence becomes unquestionable merely because a computer produced it.
That gap is where the existing law of disclosure and fair trial becomes important.
The First Distinction: The AI Tool Is Not Necessarily the Evidence
This distinction should be kept clear from the beginning.
Suppose police use AI merely to arrange 50,000 documents according to date. The software itself may have little evidentiary significance.
The position changes if the system is used to identify a particular person, classify material, detect a supposed connection between individuals, or generate an analysis that investigators subsequently rely upon.
In the second situation, the defence has a much stronger reason to ask what material was examined and what was actually produced.
So the question should not simply be:
“Was AI used?”
It should be:
“Did the AI-assisted process materially affect the evidence being used against the accused?”
What Does the BNSS Provide?
Section 230 of the BNSS requires the Magistrate to furnish the accused with the police report and specified accompanying documents in cases instituted on a police report. It includes the FIR, relevant witness statements, confessions or statements where applicable, and documents or relevant extracts forwarded with the police report.
This matters for AI-assisted investigation because the law already deals with the problem of large quantities of information.
The existence of an AI system does not, by itself, create an additional disclosure category.
Instead, the ordinary disclosure question remains:
Is the material part of the prosecution’s case, or is it otherwise relevant to a fair opportunity of defence?
The Supreme Court’s Approach to Disclosure
The Supreme Court’s earlier decisions provide a useful framework even though they were not about artificial intelligence.
In V.K. Sasikala v. State, the Court examined the accused’s request concerning documents that had not been exhibited or relied upon in the ordinary manner. The judgment recognised the prosecution’s obligation of fair disclosure while also making clear that the accused does not possess an unlimited right to every document in the police file.
That balance is particularly useful for AI.
If investigators use an AI system, the defence should not automatically receive every internal technical document merely because the word “AI” appears somewhere in the investigation.
But if an AI-generated report, underlying dataset or other material has a genuine bearing on the prosecution’s case, refusing access merely because it was generated or processed electronically could raise a serious fairness issue.
Manoj v. State of Madhya Pradesh: Why Unused Material Matters
The Supreme Court’s decision in Manoj v. State of Madhya Pradesh is another important reference point.
The Court discussed the prosecution’s disclosure obligations and the significance of material collected during investigation that could have a bearing on the accused’s defence. It also reiterated that the accused’s right to disclosure is not an unrestricted right to inspect the entire police file.
Applied cautiously to AI-assisted investigation, the principle suggests something important.
An investigator should not be able to select only the AI output that supports the prosecution and treat everything else produced during the process as irrelevant without proper consideration.
For example, if an automated analysis produces results both supporting and contradicting an investigative theory, the existence of the contradictory material may itself become relevant.
That does not mean every intermediate computer output must automatically be disclosed. It means relevance cannot disappear simply because the material was generated by a machine.
Electronic Evidence Under the BSA
The Bharatiya Sakshya Adhiniyam, 2023 gives electronic and digital records a recognised place within the law of evidence.
Section 61 states that an electronic or digital record cannot be denied admissibility merely because it is electronic or digital, subject to Section 63. Sections 62 and 63 deal with the evidentiary requirements applicable to electronic records.
This is significant, but it does not mean that an AI-generated conclusion automatically becomes reliable evidence.
There are two different questions:
- Can the electronic material legally be introduced into evidence?
- How much weight should the court give that material?
An AI system may assist in producing or analysing digital information, but the court still has to examine the evidentiary foundation of what is placed before it.
Technology can process information. It cannot remove the requirement of proof.
What Exactly Can the Accused Ask For?
The phrase “AI disclosure” can mean several different things, and they should not be treated as the same.
If an AI-generated report or output is relied upon by the prosecution, the accused has a stronger basis for seeking access to that material. The same may apply to the underlying electronic records on which the output was based, particularly where those records are relevant to challenging the prosecution’s case.
The position is less straightforward when the request concerns the source code, proprietary algorithm, model weights or complete technical architecture of the AI system. There is no general rule under Indian criminal procedure that automatically gives the accused access to such material merely because the system was used during investigation.
Similarly, the accused cannot claim every internal AI output simply because it exists. Its relevance to the prosecution’s case and its importance to the accused’s defence would matter.
The distinction can therefore be stated simply:
The use of AI may strengthen a request for disclosure of relevant evidence, but it does not automatically create a right to obtain the AI technology itself.
This distinction is particularly important because the objective of disclosure is to ensure a fair opportunity to defend, not to give the accused unrestricted access to every investigative resource.
A Useful Precedent: P. Gopalkrishnan v. State of Kerala
The Supreme Court’s decision in P. Gopalkrishnan v. State of Kerala demonstrates that disclosure can become complicated when the accused’s interests come into contact with another person’s privacy.
The case concerned electronic material and the accused’s request for access to it. The Court considered the relationship between the accused’s fair-trial rights and the privacy and dignity interests of the victim.
This offers an important lesson for future AI cases.
Even where the accused establishes a legitimate need for AI-related material, the answer may not always be an unrestricted copy of everything.
A court could instead consider inspection, controlled access, redaction, confidentiality measures or other safeguards depending on the material involved.
What If the Algorithm Makes a Mistake?
This is where judicial scrutiny becomes especially important.
An AI system may produce an incorrect classification or an unreliable association. Its output may look objective simply because it appears on a computer screen.
But computer-generated does not mean infallible.
If the prosecution relies substantially upon an AI-assisted conclusion, the defence should be able to question the factual material behind that conclusion and, where legally necessary, challenge its reliability.
The court should ultimately decide whether the evidence proves the relevant fact.
The algorithm does not become a witness merely because investigators used it.
Is There Currently a Specific “Right to AI Disclosure” in India?
No, not as a recognised standalone statutory right.
That point should be stated clearly.
There is presently no general provision in the BNSS or BSA saying that whenever an investigative authority uses artificial intelligence, the accused shall receive the algorithm or source code.
Therefore, an article claiming that Indian criminal law already guarantees such a right would go beyond the present authorities.
What can reasonably be argued is narrower:
Existing disclosure and fair-trial principles may require access to AI-generated or AI-processed material where that material is relevant to the prosecution’s case or necessary for an effective defence.
That is an argument based on existing law , not a claim that the Supreme Court has already created an AI-specific doctrine.
The Problem India Will Eventually Have to Confront
The greater difficulty may arise when AI becomes more sophisticated.
If an investigator cannot explain why an algorithm identified a particular person as a suspect, or cannot reproduce the process through which an important result was generated, traditional ideas of evidentiary transparency may be tested.
At that point, courts may have to consider questions such as:
- Who is responsible for verifying an AI-generated result?
- Can an opaque algorithm form the basis of investigative action?
- What level of technical information is necessary for effective cross-examination?
- How should confidential algorithms be balanced against the accused’s defence rights?
- What happens when the prosecution itself cannot explain why the system produced a particular result?
Indian law does not yet provide comprehensive answers to all of these questions.
That is precisely why the issue deserves attention before AI becomes deeply embedded in criminal investigations.
Conclusion
The accused does not presently have a blanket right to demand the prosecution’s AI software, source code or model.
But that does not place AI-assisted evidence beyond scrutiny.
The BNSS already provides a framework for disclosure of prosecution documents, while the BSA recognises electronic and digital records within the law of evidence. Supreme Court decisions such as V.K. Sasikala, Manoj and P. Gopalkrishnan show that disclosure, relevance, fair trial and competing privacy interests must be considered together.
The future legal position should therefore avoid both extremes.
AI should not become a reason to disclose every piece of confidential technology used during an investigation. At the same time, the prosecution should not be permitted to rely on technological conclusions while making the underlying relevant material impossible for the accused to challenge.
The real right is not necessarily a right to the algorithm. It is the right to a fair opportunity to question the evidence that the algorithm helped produce.
Frequently Asked Questions
1. Can an accused demand the source code of an AI system used by police?
Not automatically. Indian criminal law does not presently recognise a general right to obtain the source code of every investigative AI system.
2. Can the accused challenge AI-generated evidence?
Yes. The accused can challenge prosecution evidence through the ordinary procedural and evidentiary safeguards available under criminal law. The fact that technology was involved does not make the evidence immune from scrutiny.
3. Is AI-generated material automatically admissible?
No. Electronic or digital material must satisfy the applicable requirements under the Bharatiya Sakshya Adhiniyam, 2023.
4. Does the BNSS give the accused access to prosecution documents?
Section 230 provides for supply of specified prosecution documents to the accused in proceedings instituted on a police report, subject to the statutory framework and exceptions.
5. Can the accused ask for material that the prosecution did not rely upon?
There is no unlimited entitlement to the entire police file. However, Supreme Court jurisprudence recognises circumstances in which relevant material obtained during investigation may have to be disclosed to protect a fair trial.
6. Has the Supreme Court already recognised a specific “right to AI disclosure”?
No. That would be an overstatement of the present law. The issue must presently be approached through existing principles concerning disclosure, evidence and fair trial.
References
- Bharatiya Nagarik Suraksha Sanhita, 2023 Sections 230–231.
- Bharatiya Sakshya Adhiniyam, 2023 Sections 61–63.
- V.K. Sasikala v. State, (2012) 9 SCC 771.
- P. Gopalkrishnan v. State of Kerala.
- Manoj v. State of Madhya Pradesh.
- Constitution of India Articles 14 and 21.
Author: Rajnandini Verma
Rajnandini Verma ⚖️ Legal Professional & Published Writer Vivekananda College of Law, Aligarh, UP Contributing to JLRJS • Record of Law • Lawvaani Founding Member at Lawvaani | Research & Drafting IG: @rajnandiiniverma