INTRODUCTION
Prisons are a critical part of the criminal justice system, and are responsible not only for the confinement of individuals but also for their rehabilitation, reform, and reintegration into society. Even after such heavy responsibilities are put on the prisons, the prevalent conditions of Indian prisons reveal an institution strained far beyond its designed capacity. A report published by the National Crime Records Bureau (NCRB) states that as of 31 December 2023, the Indian prisons housed 5,30,333 inmates against an authorised capacity of 4,36,266, which clearly reflects an overcrowding rate of 120.8%.1 Such congestion in prisons magnifies the vulnerabilities faced by the inmates, such as psychological stress, poor living conditions, and violent behaviour, which also restricts access to medical and mental-health care.2 As per this report, 9,095 inmates (approximately 1.7%) of the total number were reported to be battling mental illness; however, because of the persistent stigma attached to psychological illness and the limited screening of such illness, this results in the underreporting of the data.3 The report also states that there were 150 deaths (approximately 7.6%) due to unnatural causes in 2023, and out of these, 96 deaths were caused by suicide, accounting for 64% of cases.4 The data clearly portray that suicide rates among prisoners are far higher than it is prevalent in the general population, which reflects the severity of psychological pressures that incarceration produces in inmates. The most prevalent contributing factors for such a scenario could include prolonged undertrial detention, separation from the family, absence of purposeful activity among inmates and inadequate medical support, especially to those who are suffering from mental-health issues.5
Under the supervision of CJI DY Chandrachud, the Supreme Court of India’s Centre for Research and Planning released a landmark report in November 2024, which urged the cautious adoption of advanced technologies to address deficiencies that are faced by the Indian prisons.6 This report emphasises that, with technological advancements worldwide, meaningful prison reform can only be achieved by incorporating enhanced mental-health monitoring systems, which are capable of supporting overburdened prison staff and also help improve inmate welfare.
With the present scenario in India with respect to overcrowding, understaffing, and rising mental health issues among prisoners, there is a concern that AI-based behavioural analysis could be one of the major supplementary tools for early detection and intervention. Globally, AI-driven behavioural analysis systems have been adopted, for instance in Austria, to identify subtle indicators of self-harm, aggression, or acute psychological distress by analysing body movements, facial micro-expressions, patterns of social withdrawal, and vocal cues.7 It could be a possibility that these technologies hold the serious potential to support prison staff by raising an early flag about high-risk situations that may otherwise go unnoticed in resource-constrained environments. However, this system can raise profound legal, ethical, and constitutional questions, particularly in the present scenario where autonomy is limited, and surveillance is a defining feature of daily life. Hence, continuous algorithmic monitoring, if implemented without robust legal safeguards, risks undermining dignity, deepening feelings of vulnerability, and misinterpreting behaviour shaped by cultural, social, or psychological diversity. Nevertheless, given the prevalent situation of the mental-health crisis in India and the limitations of the existing practices, AI-based behavioural analysis may offer important benefits if implemented within a rights-respecting framework grounded in transparency, oversight, accountability, and the fundamental humanity of those in custody.
AI AND BEHAVIOURAL ANALYSIS OF INMATES: TECHNICAL SPECIFICATION
An AI-centred behavioural analysis system heavily depends on large volumes of multimodal data, including facial feeds, audio inputs, textual communications, biometric signals and also the emotional expressions. The use of CCTV footage to analyse the behavioural irregularities among prisoners is prevalent in prison environments, which could be analysed by the prison staff through the activities of the prisoners, such as prolonged isolation, abrupt inactivity or unusual posture, which can help in identifying the potential self-harming inmates.8 However, it could take analysis of the CCTV image or video, and that could sometimes cause a delay in identifying the abnormality. With the advancement of computer vision technologies, especially in the case of convolutional neural networks (CNNs), the analysis of the visual cues to identify the patterns that could result in suicidal ideation or escalating aggression becomes easy.9 This system typically operates on the model of machine learning, where machines learn from historical data or predefined events or behaviour patterns, identify risk markers and generate predictive scores. Here, it is pertinent to note that these high-risk scores are not meant to replace human decision-making, rather to support prison staff, counsellors and medical professionals so that they can respond to the subtle change in patterns of the prisoner which could not be possible for the human to detect with naked eyes and it will eventually help them in taking efficient, quick and rational decision about the condition of the mentally ill prisoner.
A novel approach to recognising and analysing the individual’s emotion and cognitive states, such as visual cues, brain activity and, where available, speech patterns, precisely without any delay, requires integrating electroencephalography (EEG) data with AI, which works on deep learning techniques.10 Thus, it could be stated that EEG-based models, processed through 3DCNNs and combined with facial expression analysis, allow for more precise detection of internal distress, even when outward expressions are suppressed. Yet, the effectiveness of such a system in the prison context depends on the feasibility of collecting such data ethically and consistently, because a significant challenge in this regard is overcrowded, resource-constrained Indian prisons. Usually, such a system works on foreign datasets, and India, being a socially, culturally and linguistically rich country, also has a diverse crowd inside the prisons, and thus, analysing such prisoners’ behaviour requires localised data; only then can such AI-based behavioural analysis accurately predict the prisoners’ behaviour otherwise it could lead to algorithmic bias which may result in misinterpretation of ordinary behaviour.
Further, such an analysis raises not only legal but also ethical concerns because such usage of inmates’ data could lead to a violation of privacy, as consent could be taken without informing them about the use of their personal data, as they generally lack autonomy when imprisoned.11 Also, in extreme cases, AI-generated risk assessments could be misused for punitive control, such as enhanced surveillance or solitary confinement, rather than therapeutic support.12
Misclassification may worsen psychological distress instead of relieving it. Therefore, while AI offers powerful tools for early detection and intervention, its use in prisons must be carefully regulated, culturally grounded and strictly aligned with rehabilitative goals to avoid causing harm under the guise of protection.
INDIAN STATUTORY CONTEXT AND KS PUTTASWAMY THREEFOLD TEST
The 9 judges, unanimously in the landmark judgement of K.S. Puttaswamy v. Union of India13, held that violation of privacy is an integral part of a human being and it is implicit under Article 2114, and violation of it could only be allowed if the State fulfil the requirement of the threefold test: proportionality, necessity and legality.15
A. Proportionality
The test of proportionality depends upon the question of whether the intervention by the State is in the public interest or whether it is necessary and suitable for achieving the aim and objective of the policy proposed.16 It could be said that the proportionality test basically works upon the principle of balancing the interests of the State and the individual. Thus, it could be said that if the advantages resulting from such a policy measure are more in favour of the individuals than the disadvantages of it, then the test could be said to be passed. Consequently, the use of the system of AI-based behavioural analysis could only be considered proportionate if there were no suitable alternate measures available with the same level of accuracy and efficiency that could be used by the State’s stated objective of suicide prevention and custodial safety.
B. Necessity
Necessity is the mother of invention. The question of the necessity of such a system of behavioural analysis is based upon the actual need for the intervention in the mental privacy of the prisoners and possible (less intrusive) alternatives. The principle of minimal impairment demands that the State must justify why such technologically entrenched structural reforms, which directly address the causes of both violence and suicide in prisons, cannot be achieved by the same ends with fewer privacy costs. The various studies have shown that the root cause of the devastating conditions of the prisons and of the prisoners is overcrowding, stagnant living conditions, scarcity of resources and insufficient psychosocial care. However, it is possible that such a situation can be done away with by enhancing the mental-health staff, reducing the isolation of the practices, improving cell conditions and by most importantly, by creating a trusted relationship between staff and inmates. But here the question arises whether this is sufficient for the reduction of suicide and to tackle mental health issues; the answer to that could be negative. Therefore, AI-based monitoring of inmates comes into the picture, which addresses the symptoms rather than the causes; its deterrent value in the context of spontaneous or displaced violence remains largely speculative.17 It is pertinent to understand that the State cannot opt for a minimal-impairment standard by pointing to administrative convenience or resource scarcity; the K S Puttaswamy judgement makes it clear that rights cannot be given way simply because invasive measures are easier to implement.18 Unless the State evidences that structural reforms or human-centred interventions are incapable of reasonably achieving the protective aim, AI cannot be justified as necessary.
C. Legality
No policy measure of a State infringing the privacy of an individual can be said to be legal if it lacks the principle of legality, which means legality is the gateway requirement for any law to survive the constitutional scrutiny and to be declared valid and not void.
At present, India lacks any statutory law that authorises AI-based behavioural analysis of prisoners. The Prisons Act of 189419 and the Model Prison Manual 201620 are silent on algorithmic monitoring, and the recently enacted Digital Personal Data Protection Act, 202321, also does not provide a framework for processing the extraordinarily sensitive behavioural and psychological data generated by such systems. Although Section 45 of the Model Prisons and Correctional Services Act, 202322, acknowledges the centrality of psychological assessment and treatment and aims to reform prison systems through modern humane approaches; however, the Implementation has been uneven and insufficient.23,24 The Model Prisons & Correctional Services Act, 2023 provides a model framework for State adoption of prison administration and correctional services reforms, and it explicitly references the Mental Healthcare Act, 201725, in particular Section 10326 (addressing ‘prisoners with mental illness’) and the broader right to mental healthcare under Section 1827, thereby obliging States to integrate mental-healthcare provisions for inmates into the prison system. The lesson could be learnt from the European Artificial Intelligence Act (AI Act)28, one of the most comprehensive AI Acts in the world, which allows such intervention in extraordinary situations.29
Beyond legislative authorisation of a statute, the principle of legality also requires that the enacted law is based upon the principle of Natural Justice and does not violate fundamental rights of the individuals without due cause, thus, the purpose of it, the remedy in case of violation should be precisely mentioned without any ambiguity then only the statute could be said to be valid in the eyes of the law. The Supreme Court, in a landmark case of Selvi v State of Karnataka30, further underscores the principle of mental privacy and held that the nonconsensual or forceful extraction of mental content violates both the dignity and autonomy of an individual, thus upholding the unconstitutionality of the Narco test, Polygraph test, etc.31 Therefore, any technology capable of inferring emotional or psychological states must be subject to the most rigorous legal constraints and must not fall outside the legal boundaries of the Constitution of India.
ETHICAL PERSPECTIVES: UTILITARIANISM AND DEONTOLOGY
The ethical debate around AI-based behavioural analysis in prisons turns on a central question: not whether such systems are technologically capable, but whether they ought to be used at all in a space defined by coercion, vulnerability, and limited autonomy.
The utilitarianism theory32, consequential in nature, is based on the principle of the greatest good at the lowest cost for the community, that is possible net benefit is evaluated.33 If this perspective is adopted, then the AI-based behavioural analysis could be adopted, which not only lowers the burden of the prison staff but at the same time increases the safety of the prisoners, thus ethically justifying the use of such technology. Thus, the social approach is taken over by a “new penology” approach, which is based on technological surveillance. It means that the rehabilitation and the prospect of the (re)integration of marginalised individuals seem to play a subordinate role.34 In this sense, one could argue that the deployment of the AI system in question is not about the ‘individuals in need of treatment’, but rather about automated ‘techniques to identify, classify and manage’ risks associated with the detained subjects. AI thus appears as a technological extension of Bentham’s panopticon, optimising surveillance for institutional safety.
In a contrasting moral lens compared to the utilitarian theory, a deontological approach emphasises the personal rights of an individual, which include human dignity and autonomy. This theory is grounded in Kantian ethics and echoed in Supreme Court jurisprudence from Sunil Batra35 to Common Cause36. It holds that the moral status of an action is not based on the consequence of it, but rather upon how it is applied.37 Every individual possesses intrinsic dignity and cannot be treated merely as a means to administrative ends, and such dignity is not only available to every individual, even if they are behind bars. Thus, from the standpoint of the deontological theory, it could be said that the use of AI-based surveillance should not be implemented, as it violates the privacy of the prisoners. However, on the other hand, this approach also considers any form of voluntary subjugation, intentional self-mutilation or suicide as morally wrong. Hence, it can be concluded that such surveillance may be morally impermissible unless it serves a narrowly defined and compelling duty to protect life and dignity.
CONCLUSION
For India to adopt responsible AI-based behavioural analysis in prisons, the first step in that direction must begin with framing a clear legal and ethical policy. Any technology that deals with private data of an individual must fulfil the fundamental requirement of the Constitution of India; thus, without explicit legislative backing, such a system should be held unconstitutional. The statute, if enacted for AI-based analysis, must define the proper purpose and should be limited to the prevention of suicide and for mental-health support, and it must also define how data may be collected, how long it can be stored, who can access it, and what penalties may be applicable in case of a breach. These safeguards should not be optional; rather, strict adherence should be made. The threefold test of legality, legitimacy, and proportionality, which evolved in the case of K S Puttaswamy, should also be strictly followed. The well-informed consent of the inmates should be obtained wherever possible, and also a strong data protection measures, such as regular independent audits, secure channels, and anonymisation, should be implemented.
Ultimately, it could be said that overcrowding, understaffing, and the high incidence of inmate suicides make it clear that the system needs better tools for early intervention. Yet technology alone cannot cure structural injustice. It should be noted that the technology should not be seen as a replacement; rather, it should be used as a supporting tool. Thus, AI-based behavioural analysis cannot replace clinical judgment but should instead support mental-health professionals and prison staff in recognising early signs of distress that human monitoring might miss; it should only be for rehabilitative and protective purposes, not as a tool for stringent punitive surveillance.
*****
Footnotes
1. National Crime Records Bureau, Ministry of Home Affairs, ‘Prison Statistics India – 2023’ (2025).
2. Shreehari Paliath, ‘‘Jail Within A Jail’: Death Row And The Impact On Mental Health’ (IndiaSpend, 9 October 2025) <https://www.indiaspend.com/governance/jail-within-a-jail-death-row-and-the-impact-onmental-health-968507> accessed 20 November 2025.
3. Shreehari Paliath, ‘The Brewing Mental Health Crisis In Indian Prisons’ (IndiaSpend, 2 August 2025) <https://www.indiaspend.com/governance/the-brewing-mental-health-crisis-in-indian-prisons-963116> accessed 20 November 2025.
4. Ambika Pandit, ‘1.7% of jail inmates in 2023 faced mental health issues: NCRB report’ The Times of India (India, 24 October 2025) <https://timesofindia.indiatimes.com/india/1-7-of-jail-inmates-in-2023-faced-mentalhealth-issues-ncrb-report> accessed 19 November 2025.
5. Neha Kumari and Madhukar Anand, ‘Mental Health in Indian Prisons: A Legal Perspective with Comparative Insights from the Netherlands’ (2025) 8 International Journal of Law Management & Humanities 2560.
6. Centre for Research and Planning, Supreme Court of India, ‘Prisons in India: Mapping Prison Manuals and Measures for Reformation and Decongestion’ (2024).
7. Robert Rothmann and Elisabeth Mayer, ‘AI-Based Behaviour Analysis in Prisons: A Multidisciplinary Assessment of an Austrian Case Study’ (2025) 22 European Journal of Criminology 1.
8. Troy J Allard, Richard K Wortley and Anna L Stewart, ‘The Effect of CCTV on Prisoner Misbehavior’ (2008) 88 The Prison Journal 404.
9. Julian Strohmayer and Martin Kampel, ‘A Compact Tri-Modal Camera Unit for RGBDT Vision’ (5th International Conference on Machine Vision and Applications, Singapore, February 2022).
10. Anupama Budhewar and others, ‘Human Behaviour Analysis Using CNN’ (2024) 194 SHS Web Conf 01001.
11. Veronika Hofinger and Philipp Pflegerl, ‘A Reality Check on the Digitalisation of Prisons: Assessing the Opportunities and Risks of Providing Digital Technologies for Prisoners’ (2024) 26 Punishment & Society 898.
12. Justin Iverson, ‘Surveilling Potential Uses and Abuses of Artificial Intelligence in Correctional Spaces’ (2022) 9 Lincoln Memorial University Law Review 1.
13. K.S. Puttaswamy v. Union of India, (2019) 1 SCC 1.
14. The Constitution of India, 1950, art 21.
15. supra note 14.
16. supra note 08.
17. supra note 08.
18. supra note 14.
19. The Prisons Act, 1894 (Act IX of 1894).
20. The Model Prison Manual 2016.
21. The Digital Personal Data Protection Act, 2023 (Act 22 of 2023).
22. The Model Prisons and Correctional Services Act, 2023, s 45.
23. ‘Mental Health & Criminal Law’ (Mariwala Health Initiative) <https://mhi.org.in/voice/details/mentalhealth-criminal-law/> accessed 20 November 2025.
24. Rajnandini Rathod, ‘What is the Mental Health Act 2017’ (Rehabs.in) <https://rehabs.in/news/what-is-themental-health-act-2017/> accessed 20 November 2025.
25. The Mental Healthcare Act, 2017 (Act 10 of 2017).
26. id., at s 103.
27. id., at s 18.
28. ‘EU AI ACT: First Regulation on Artificial Intelligence’ European Parliament (EU, 19 February 2025) <https://www.europarl.europa.eu/topics/en/article/20230601STO93804/eu-ai-act-first-regulation-on-artificialintelligence> accessed 20 November 2025.
29. Mike Nellis, ‘Artificial Intelligence and Correctional Practice: The Council of Europe Draft Recommendation on the Use of AI in Prisons and Probation’ (2024) 37 The Journal of Offender Monitoring 11.
30. Selvi v. State of Karnataka, (2010) 7 SCC 263.
31. ibid.
32. Henry R. West, An Introduction to Mill’s Utilitarian Ethics (Cambridge University Press 2003).
33. Richard A Spinello, Cyberethics: Morality and Law in Cyberspace (7th edn, Jones and Bartlett Publishers 2021).
34. Malcolm M Feeley and Jonathan Simon, ‘The new penology: Notes on the emerging strategy of corrections and its implications’ (1992) 30 Criminology 449.
35. Sunil Batra v. Delhi Administration, (1978) 4 SCC 494.
36. Common Cause v. Union of India, (2018) 5 SCC 1.
37. Mary Gregor and Jens Timmermann (trs), Groundwork of the Metaphysics of Morals (Cambridge University Press 2011).