Author: Shavi Sharma, Chanderprabhu Jain College
Abstract
Artificial Intelligence (AI) is revolutionising the criminal justice system by improving the efficiency of crime prevention, criminal investigations, evidence analysis, and judicial administration. AI-powered technologies have significantly enhanced the ability of law enforcement agencies to detect and investigate criminal activities. However, the increasing reliance on AI in criminal justice also raises complex legal, ethical, and constitutional concerns. The absence of accountability, lack of transparency, and data privacy violations may weaken the principles of fairness, equality, and due process guaranteed under the Constitution of India.
This article evaluates how artificial intelligence fits — or fails to fit — within India’s constitutional and statutory criminal justice architecture, examining its alignment with Articles 14, 20, 21, and 22, and with the Bharatiya Nyaya Sanhita, Bharatiya Nagarik Suraksha Sanhita, Bharatiya Sakshya Adhiniyam (all 2023), the Information Technology Act, 2000, and the Digital Personal Data Protection Act, 2023. It examines whether AI-generated digital evidence can be admitted in court, the risks predictive policing poses to individual liberty, and why judicial discretion must remain central to criminal adjudication
AI can undoubtedly make criminal justice faster and more consistent — but it cannot replace human judgment or constitutional reasoning. It should assist judges and investigators, not decide cases on their behalf. For India to use AI responsibly, it needs a clear legal framework built on transparency, accountability, and human oversight — one that keeps technology answerable to the Rule of Law and Fundamental Rights, rather than the other way around.
To the point
AI is now active across Indian policing, bail, sentencing, and forensic evidence this article tests whether that use holds up against the constitution and current criminal law.
The use of AI in criminal justice implicates four constitutional guarantees: Article 14 (equality before law), Article 20 (protection against self-incrimination and double jeopardy), Article 21 (right to life and personal liberty, encompassing privacy), and Article 22 (safeguards against arbitrary arrest and detention).
The new criminal law trio — BNS, BNSS, BSA (2023) — along with the IT Act, 2000 and the DPDP Act, 2023, forms the statutory backdrop AI tools must operate within, even though none was designed to address AI.
A key open question: can AI-generated digital evidence be admitted in court under the Bharatiya Sakshya Adhiniyam’s evidentiary standards, and who is accountable when that evidence is wrong?
Predictive policing raises the same profiling and bias risks in India as elsewhere — data-driven “risk” flagging before any offence has occurred sits uneasily with the presumption of innocence.
Core argument: AI should remain a decision-support tool, never an autonomous decision-maker — judicial discretion and reasoning cannot be delegated to a machine.
What’s missing in India right now: a dedicated legal framework requiring transparency, accountability, and human oversight for AI used in criminal justice — the article argues this is essential to keep innovation aligned with the Rule of Law and Fundamental Rights.
Use of legal Jargon
This section sets out key legal terms central to Indian constitutional and Criminal jurisprudence that recur throughout the article.
Rule of law: AI in criminal justice must always work under the authority of law, with transparency, accountability, and judicial control — not according to the unchecked discretion of technology or government officials.
Natural Justice: The twin principles of audi alteram partem (the right to be heard) and nemo judex in causa sua (rule against bias), both of which are strained when an algorithmic system influences a decision without disclosing its reasoning or permitting the affected person to respond to it.
Due Process of Law: Read into Article 21 through judicial interpretation, requiring that any procedure depriving a person of life or liberty be fair, just, and reasonable — a standard difficult to satisfy where AI-generated risk scores or evidence are not independently explainable.
Self-incrimination (Article 20(3)): The constitutional bar against compelling an accused to be a witness against themselves, relevant where AI-assisted interrogation or behavioural-analysis tools risk extracting involuntary disclosures.
Admissibility and Evidentiary Value: Under the Bharatiya Sakshya Adhiniyam, evidence must satisfy tests of relevance and reliability; AI-generated digital evidence raises fresh questions of authentication, chain of custody, and expert scrutiny.
Predictive Policing: The use of algorithmic profiling to anticipate criminal activity, implicating both the presumption of innocence and protection against arbitrary arrest under Article 22.
Judicial Discretion: The irreducible core of adjudication whereby a judge exercises independent reasoning; the article’s central argument is that this discretion cannot be delegated to, or substituted by, an automated system.
The proof
A. Predictive Policing
Several Indian police departments are now turning to data-based crime mapping to guide patrol deployment. The concern is simple: if the historical arrest data being fed into these systems already reflects skewed or excessive policing of particular communities or localities, the algorithm doesn’t correct for that bias — it absorbs and repeats it, directing patrols back to the same areas regardless of actual crime levels. What looks like an objective prediction is often just yesterday’s bias, automated.
B. AI in Bail and Sentencing
Indian courts are gaining access to tools that produce risk scores to assist bail and sentencing decisions. The difficulty is that these scores frequently come from systems whose internal logic remains undisclosed, leaving both judges and the accused unable to meaningfully question how a particular “risk” figure was arrived at. This opacity sits uneasily with the fairness guarantee under Article 21.
C. Facial Recognition and Surveillance
A number of Indian police forces have adopted facial recognition to identify suspects captured on CCTV. Studies from around the world have repeatedly found that such systems misidentify women and individuals with darker skin tones at disproportionately higher rates. Applied in India, this raises two distinct concerns — an equality problem under Article 14, and a privacy problem under Article 21, given that Puttaswamy recognized privacy as a fundamental right.
D. AI-Assisted Forensic Evidence
Digital forensics — extracting and interpreting data from phones, laptops, and other devices — increasingly relies on AI tools. The Bharatiya Sakshya Adhiniyam requires that such evidence meet standards of reliability and authenticity before being admitted. Yet when the underlying AI system is proprietary or its methodology unclear, the defense is left with little real ability to challenge how the evidence was produced — weakening the accused’s right to contest it meaningfully.
Overall Point
A common thread runs through all four areas: tools introduced to improve efficiency end up creating fairness gaps that existing law was never built to address. Indian criminal law, as it stands, is still designed around a human decision-maker — someone whose reasoning can be seen, questioned, and reviewed. AI disrupts that assumption at every stage.
Case laws
1. Justice K.S. Puttaswamy v. Union of India, (2017) 10 SCC 1
Facts: The case arose from a challenge to the Aadhaar biometric scheme, questioning whether state collection of citizens’ biometric and demographic data violated fundamental rights. A threshold question was whether privacy was even a fundamental right, since earlier decisions had held it was not.
Significance: This case is foundational to any discussion of AI, surveillance, or data collection by the state, since it directly governs facial recognition, predictive policing, and biometric surveillance.
Judgment: A nine-judge bench unanimously held that privacy is a fundamental right, intrinsic to Article 21. The Court laid down a threefold test for state action infringing privacy: legality (backed by law), necessity (a legitimate aim), and proportionality (proportionate means). Any AI surveillance tool used by Indian law enforcement must now satisfy this test.
2. Selvi v. State of Karnataka, (2010) 7 SCC 263
Facts: The case concerned involuntary administration of narco-analysis, polygraph tests, and brain electrical activation profiling on accused persons during investigation, without consent, to extract information or confessions.
Significance: This judgment is significant for AI-assisted interrogation and behavioural-analysis tools, since it limits how far investigative techniques can go in extracting information, even where the technology is advanced.
Judgment: The Court held that involuntary administration of these techniques violates Article 20(3) and Article 21. Such tests can only be conducted with free, informed consent, and even then the results cannot be used as substantive evidence of guilt — a strong precedent against coercive use of technology in investigation.
3. Maneka Gandhi v. Union of India, (1978) 1 SCC 248
Facts: The petitioner’s passport was impounded without reason or a hearing. She challenged this as a violation of her right to personal liberty under Article 21.
Significance: This case is central to evaluating AI-based bail, sentencing, and risk-assessment tools, since it defines what “fairness” means when the state deprives someone of liberty — a standard opaque algorithms struggle to meet.
Judgment: The Court held that “procedure established by law” under Article 21 must be fair, just, and reasonable — not arbitrary. Reading Articles 14, 19, and 21 together, it held any law depriving liberty must pass a reasonableness test. This means an AI tool influencing bail or sentencing must itself be explainable and contestable.
Conclusion
Artificial intelligence is no longer a distant possibility for India’s criminal justice system — it is already shaping policing, bail, sentencing, and forensic evidence. Used well, it can ease the burden on an overstretched system, bringing greater speed and consistency to decisions that currently suffer from delay and disparity. But this article has shown that efficiency cannot come at the cost of constitutional guarantees. Article 14 demands that AI tools not entrench bias under the guise of neutrality. Article 20 demands that no technology coerce an accused into self-incrimination. Article 21, as Puttaswamy and Maneka Gandhi establish, demands that any process affecting life or liberty be fair, transparent, and reasoned — a standard that opaque, unexplainable algorithms cannot easily meet. And Article 22 demands that no one be arrested or detained on the basis of a machine-generated risk score alone.
India’s new criminal law framework — the BNS, BNSS, and BSA — along with the IT Act and the DPDP Act, was not written with AI in mind, and this gap is the central problem this article has identified. Predictive policing, algorithmic bail tools, facial recognition, and AI-assisted forensic evidence all operate today in a legal grey zone, tested only indirectly by precedents like Selvi, Bridges, and Loomis rather than by any dedicated Indian statute.
The way forward is not to reject AI, but to discipline it. AI must remain a decision-support tool — assisting judges, investigators, and forensic experts — never an autonomous decision-maker substituting for judicial reasoning. This requires a dedicated legal and regulatory framework built on four pillars: transparency in how these systems function, accountability for their errors, mandatory human oversight at every stage, and a genuine right for the accused to contest algorithmic findings in court. Only such a framework can ensure that as India modernizes its criminal justice system, it strengthens — rather than compromises — the Rule of Law, Natural Justice, and the fundamental rights that anchor its constitutional democracy.
FAQs
Q1. Is AI currently used in Indian courts or police forces?
Yes. Several Indian states use AI-based facial recognition for suspect identification, data-driven crime mapping for patrol deployment, and digital forensic tools for evidence analysis. However, there is no dedicated Indian statute yet regulating AI’s use specifically in criminal justice.
Q2. Does the Indian Constitution directly mention AI?
No. Articles 14, 20, 21, and 22 were not drafted with AI in mind, but courts interpret them broadly enough to apply to new technologies. Their protections — equality, protection against self-incrimination, life and personal liberty, and safeguards against arbitrary arrest — extend to AI-driven state action even without express mention.
Q3. Can AI-generated evidence be used in an Indian court?
It can be, provided it satisfies the reliability and authenticity standards under the Bharatiya Sakshya Adhiniyam. The unresolved difficulty is that many AI systems are proprietary or unexplainable, making it hard for the defense to challenge how the evidence was actually generated.
Q4. Can a court in India rely solely on an AI-generated risk score to deny bail or determine a sentence?
Not safely, under current constitutional standards. Following Maneka Gandhi, any decision affecting liberty must be based on a fair and reasoned procedure. A risk score that cannot be explained or questioned would likely fail this test if challenged.
Q5. What did Puttaswamy establish that is relevant to AI?
It recognized privacy as a fundamental right under Article 21 and laid down a three-part test — legality, necessity, and proportionality — that any state surveillance or data-collection activity, including AI-based tools, must satisfy.
