Three in Four

Pritiraj BrahmaFounding Partner, Pritiraj & Partners · 12 April 2026

India’s prison population fell in 2023 and overcrowding eased by ten points. Nearly three-quarters of those inside have still not been convicted of anything.

India did not decongest its prisons by building them. It decongested them by holding fewer people — which is the right method, arrived at without ever being adopted as a policy.

The National Crime Records Bureau’s Prison Statistics India 2023, released in September 2025, contains the first substantial good news on Indian prisons in some years, and it has been almost entirely overlooked. The prison population fell from 5,73,220 at the end of 2022 to 5,30,333 at the end of 2023. The all-India occupancy rate fell from 131.4 per cent to 120.8 per cent. Undertrial numbers dropped by roughly a tenth.

The reason this matters is not the size of the improvement but its mechanism. Capacity rose by 0.7 per cent — two prisons, from 1,330 to 1,332, and sanctioned capacity from 4,36,266 to 4,39,119. Essentially none of the decongestion came from construction. It came from fewer people being held before trial. That is the lever, and it was pulled almost by accident.

Six findings anchor this analysis:

1. Overcrowding eased materially. Occupancy fell from 130.2 per cent in 2021 and 131.4 per cent in 2022 to 120.8 per cent in 2023 (NCRB, PSI 2023). Prisons still held roughly a fifth more people than they were built for.
2. Undertrials remain the overwhelming majority. 3,89,910 undertrial prisoners at the end of 2023, against 4,34,302 a year earlier — a fall of about 10 per cent, but still close to 73.5 per cent of all prisoners. Convicts rose slightly, from about 1.33 lakh to 1.35 lakh.
3. Three states hold two-fifths of India’s undertrials. Uttar Pradesh 73,491 (18.8 per cent), Bihar 46,529 (11.9 per cent), Maharashtra 32,438 (8.3 per cent). Several prisons in these states operate between 150 and 250 per cent of sanctioned capacity.
4. The national average hides the extremes. Delhi’s prisons ran at close to 200 per cent occupancy — the highest of any state or union territory — while Telangana was lowest at 72.8 per cent. A 120.8 per cent average describes almost no actual prison.
5. Pre-trial detention falls disproportionately on the same groups. Roughly two in three undertrials belong to Scheduled Castes, Scheduled Tribes or Other Backward Classes, and nearly half are aged between 18 and 30 (IndiaSpend’s analysis of PSI 2023).
6. The data arrived two years late. PSI 2023 was published in September 2025 — two years after the preceding edition, and describing a position as on 31 December 2023.

One caveat, stated plainly: secondary compilations of PSI 2023 do not fully agree. At least one widely circulated summary reports a total prison population of 5,54,034 and an undertrial share of 76.2 per cent, against the 5,30,333 and 73.5 per cent reported by NCRB and by the news organisations that worked from the published report. We use the latter, and flag the discrepancy rather than resolve it silently — it is itself evidence of how difficult this dataset is to use.

The number that fell, and why

Undertrial Review Committees were established at district level following a 2015 Supreme Court order, charged with periodically identifying prisoners eligible for release — those who have served the maximum sentence for their alleged offence, those eligible for bail they cannot furnish, those held for offences now compoundable. Their most recent published data dates from April 2025. IndiaSpend attributes part of the 2023 decline to their functioning.

If that attribution is right, it is the most important sentence in Indian prison policy in a decade, because it means the population is responsive to an administrative mechanism that already exists, costs almost nothing, and is not being measured. No one publishes, quarterly and by district, how many cases a UTRC reviewed, how many releases it recommended, and how many actually occurred. A lever that works and is not monitored will be pulled unevenly, and will stop being pulled when attention moves.

The composition tells you what the system is actually doing

A prison population that is three-quarters unconvicted is not a corrections system with a backlog. It is a detention system with a courtroom attached. The relevant statistic is not how many people are inside but how long they have been inside without trial — and that figure is not published in a usable form. PSI reports the stock of undertrials by period of detention, but not the median duration of pre-trial custody by district or offence, which is the number a district judge or a state home secretary would need to act.

The demographic pattern compounds this. When two-thirds of those held before trial come from Scheduled Castes, Scheduled Tribes and Other Backward Classes, and half are under thirty, the machinery is not distributing pre-trial liberty evenly. The most common reason a bailable prisoner remains in custody is inability to furnish surety — which is to say, poverty operating as a sentencing factor.

Data that arrives too late to govern by

A report published in September 2025 about 31 December 2023 cannot inform an operational decision. It can only inform a debate. The India Justice Report 2025 and the April 2025 UTRC data are more current, but neither is an official annual series with state-wise comparability.

This is the firm’s recurring theme in a particularly stark form: the state collects the number, does not publish it in time, and therefore cannot be held to it. Prison occupancy is recorded daily in every jail in India. There is no technical reason it is a biennial revelation.

The counter-case, honestly stated

Three points deserve weight.

First, the improvement is real and should not be argued away. A ten-point fall in occupancy and a 10 per cent fall in undertrials in a single year is a substantial change, achieved without capital expenditure. Any analysis that treats 2023 as more of the same is misreading the data.

Second, a high undertrial share is partly arithmetic, not only injustice. Undertrials cycle through custody quickly; convicts serve long terms. A system that processes cases faster will still show a large undertrial stock at any given snapshot. International comparison bears this out — India’s occupancy rate is unremarkable by global standards, even as its undertrial share is high.

Third, some pre-trial detention is necessary. Flight risk, witness intimidation and the seriousness of certain offences are legitimate grounds, and a policy conversation that treats every undertrial as wrongly held will lose the argument with those responsible for public order.

The case does not rest on abolishing pre-trial detention. It rests on the fact that nobody in the system is currently required to answer for its duration.

What we would do

1. Publish prison occupancy monthly, by prison. The data exists in every jail register. A single national page, updated monthly, showing sanctioned capacity and actual population for each of India’s 1,332 prisons would end the two-year information lag at negligible cost.
2. Make median pre-trial detention duration the headline indicator. Not the count of undertrials — the median days in custody before trial concludes, published by district. It is the number that captures the harm, and it is computable from data the courts and prisons already hold jointly.
3. Give every UTRC a published quarterly return with a named owner. Cases reviewed, releases recommended, releases effected, and reasons for the gap. The district judge and district magistrate co-chair these committees; both should sign the return. A committee whose output is published behaves differently from one whose output is not.
4. Fund surety, or stop pretending bail was granted. Where a court has granted bail and the prisoner remains inside solely for want of surety or bond money, that is a failed release, and it should be counted as one. States should report the number monthly and legal services authorities should be resourced to close it — this is the cheapest liberty India can buy.
5. Set the three-state target explicitly. Uttar Pradesh, Bihar and Maharashtra hold two-fifths of India’s undertrials. A national programme with a diffuse target across 36 jurisdictions will not move the number. Three chief secretaries, three published baselines, three annual targets.

India reduced its prison population last year without building a prison. It should now find out how, write it down, and do it deliberately.