Few figures in Indian public life are quoted as confidently, and read as carelessly, as the conviction rate under the anti-terror law. A percentage circulates on social media, stripped of its period, its source and the thing it divides by, and is deployed to prove both that the Unlawful Activities (Prevention) Act is an instrument of mass wrongful detention and that the same law is working exactly as intended. The anti-terror law enforcement record is not short of data. It is short of readers who know what each figure counts. Both uses of the circulating percentage draw on the same pool of official numbers, and both are wrong in ways that a careful reader can detect without any special access, because the error is arithmetic rather than ideological.

The argument of this article can be stated plainly, and everything that follows is the evidence for it. There are four different things that get called the record: the number of cases registered, the number of persons arrested, the number of persons charge-sheeted, and the number of persons convicted. Each is compiled by a different process, each measures something different, and each fails in a different way when it is used as the denominator for a conviction rate. The single most repeated figure in the public debate, an annual conviction percentage obtained by dividing the persons convicted in a calendar year by the persons arrested in that same calendar year, is not a conviction rate at all. It is a ratio of two flows from different cohorts, and it measures the speed of the disposal system more than it measures the accuracy of the prosecutions. That correction does not make the underlying facts less troubling. It makes them honest, and the honest version turns out to be enough to carry a serious argument about what the statute does in practice.

The reader who finishes this article will be able to do three things that almost nobody in the public debate does. First, the reader will be able to trace any quoted percentage to the parliamentary answer or the National Crime Records Bureau volume it came from, and to say which period it covers and which denominator it uses. Second, the reader will be able to spot the two standard misuses, which are the critic’s error of treating a low quotient as proof of misuse and the defender’s error of treating the time lag as an excuse that dissolves the gap. Third, the reader will be able to state, defensibly and in one sentence, what the data actually establishes about the enforcement of anti-terror law in India. The article is written in that spirit of being harder on the evidence than anyone who disagrees would be, because that is the only way enforcement data becomes an argument rather than a slogan.

A word on the boundary of this article. It is about the enforcement record of the UAPA as reported in the official statistics, and not about the substantive law of the statute, which the series covers in the complete guide to the Act. The link to that guide appears at the first point where the statute’s provisions become relevant to the figures. The companion lineage article traces the statutes that preceded the UAPA, and the bail interpretation article owns the law of pre-trial detention that explains why the figures look the way they do.

The anti-terror law enforcement record begins with its sources

Before any figure can be read, the reader needs the source map, because the Indian enforcement statistics system is not one dataset but a small archipelago of them, and each publishes only what it was designed to count. The most important source is the National Crime Records Bureau’s annual publication, Crime in India, which compiles data as reported to it by the states and union territories. The Bureau does not investigate cases, it does not verify them independently, and it does not follow individual accused persons across years. It receives returns in prescribed formats, aggregates them, and publishes the aggregate. For the UAPA, the relevant material sits in the chapters on offences against the State, where the Bureau reports cases registered under the Act, persons arrested, and the disposal of cases and persons by the courts. The Bureau’s tables use fixed abbreviations for the categories of disposal, and the Ministry’s answers in Parliament reproduce those tables verbatim, so the reader who learns the abbreviations can read the primary material directly.

The second source is the set of answers given by the Ministry of Home Affairs to questions asked in the Lok Sabha and the Rajya Sabha. These answers are the primary public source for state-wise and year-wise arrest and conviction figures, and they are what journalists and researchers actually cite, usually at one remove through news reports that strip out the period and the caveat. A typical answer reproduces a Bureau table for a stated period, adds the ministry’s own framing, and sometimes answers the policy question behind the numbers, as when the ministry told the Rajya Sabha in December 2021 that conviction is an outcome of an elaborate judicial process dependent on trial duration, the appraisal of evidence and the examination of witnesses, and that adequate constitutional, institutional and statutory safeguards exist against misuse. The ministry’s answers also record the limits of the data, and those limits are themselves important evidence, as when the same ministry told Parliament that the Bureau does not maintain data on how long individual accused persons spend in jail, which means that no official series exists for the average length of pre-trial custody under the Act.

One more official source belongs in the map, although the debate rarely uses it. The Bureau also publishes Prison Statistics India, which counts the prison population by the statute under which undertrials are held, and which therefore offers an independent check on the custody stock: the undertrial population recorded in the prisons at year end under the UAPA can be set against the judicial custody figures in the ministry’s answers, and the two series, compiled through different returns, broadly corroborate each other’s scale. The prison statistics also record the undertrial population’s share of the total prison population across all offences, a reminder that prolonged pre-trial custody is a general pathology of the Indian system within which the UAPA population is an extreme case rather than a separate phenomenon. The usable, disaggregated series for the Act in the Bureau’s crime volumes begins in the middle of the last decade, which is why every serious study in the field starts its count in 2015 or 2016: for the Act’s first four and a half decades there is no comparable published series at all, and claims about long-run trends across the statute’s whole life have no statistical foundation.

The third and fourth sources are the records that the official system does not aggregate. One is the case information published by the National Investigation Agency on its website, which a civil liberties study used alongside the Bureau volumes, and which counts a small but distinctive slice of the enforcement record, being the cases the Centre has taken over rather than the cases the state police register. The other is the body of judicial orders themselves, from the Supreme Court down to the special courts, in which individual cases leave a dated trail of arrest, charge sheet, bail applications and final orders. No published table connects the Bureau’s arrest figure to the court’s disposal figure for the same cohort of accused persons, and that missing link is the central methodological fact of this entire field. Every honest statement about the enforcement record has to be made across that gap, and every dishonest one pretends the gap does not exist.

The Bureau’s own presentation contains one more trap that even careful readers walk into, which is the meaning of the Bureau’s conviction rate. When the Bureau prints a conviction rate, it computes it as the percentage of cases in which trial was completed that ended in conviction, not as a percentage of persons arrested. For the UAPA over the period from 2015 to 2020, the People’s Union for Civil Liberties calculated the Bureau’s own case-based rate at 27.57 per cent, against 49.67 per cent for Indian Penal Code cases over the same period. The same study calculated the person-based quotient, persons convicted divided by persons arrested in each year, at 2.8 per cent, against 22.19 per cent for the Penal Code. Both numbers come from the same Bureau volumes. The first says that roughly a quarter of UAPA trials that reach a verdict end in conviction. The second says that the number of persons convicted in a year is a small fraction of the number arrested in that year. The first is a rate in the proper sense, conditional on a completed trial. The second is a ratio of two flows, and it is the second that circulates in public as the conviction rate. Keeping those two numbers apart in the mind is the whole skill this article teaches.

Who compiles the enforcement figures that Parliament and the courts quote?

The National Crime Records Bureau compiles them from returns filed by the states and union territories and publishes them in the annual Crime in India volumes, and the Ministry of Home Affairs reproduces those compilations in answers to questions asked in the Lok Sabha and the Rajya Sabha.

Courts, scholars and journalists almost always encounter the figures through one of those two channels, with the ministry’s parliamentary answers serving as the primary public source for year-wise and state-wise breakdowns. The Bureau aggregates what the states report; it does not independently verify cases or follow individual accused persons across calendar years. Understanding that chain, from the police station return to the Bureau table to the ministry’s answer to the news report, is what lets a reader check any quoted percentage against its origin instead of accepting it as a free-floating fact.

How a ministry answer is built, and how to read one

Because the ministry’s parliamentary answers are the working source for almost everyone who writes about this record, the reader benefits from knowing how such an answer is assembled and where its soft points are. A member of either House files a question, starred questions receiving an oral answer with supplementary debate and unstarred questions receiving a written reply laid on the table of the House. The question itself frames the period, and members who ask about enforcement usually ask for the figures for the preceding three or five years, state by state. The ministry does not compile the answer from its own records, because the Union government does not hold case-level data on state prosecutions. It reproduces the Bureau’s published tables for the period requested and prefixes them with a short doctrinal paragraph. The result is a document of a peculiar kind: an annexure of hard numbers carrying a gloss of institutional reassurance, in which the numbers are the most reliable part and the gloss is the part that gets quoted.

The December 2021 exchanges display the form at its most instructive. Dr V. Sivadasan’s unstarred question asked how many people had been arrested under the UAPA in five years, how many had been punished, in how many cases courts had revoked the application of the Act, how many accused had been in jail for more than a year, and how many were in custody. The answer gave the Bureau’s year-wise tables for the first three parts, recorded that the Bureau does not maintain the duration data the fourth part required, and gave the year-end custody stock for the fifth. Anand Sharma’s starred question in the same session asked for arrests and convictions year-wise and then asked whether the government planned to amend the Act in light of the evidence about large numbers of acquittals, whether it intended to prevent the harassment of innocent persons, and if not, why not. The reply to the policy parts was a single doctrinal passage: conviction is an outcome of an elaborate judicial process dependent on the duration of trial, the appraisal of evidence and the examination of witnesses, and adequate constitutional, institutional and statutory safeguards, including safeguards built into the UAPA itself, exist to prevent misuse. That passage, or close variants of it, appears in answer after answer across the years, and recognising it as boilerplate is the beginning of reading these documents critically. The annexure changes with the data. The reassurance does not change with anything.

Three reading rules follow from the way the documents are built. The first is to read the annexure and discount the gloss: the tables are the Bureau’s published data and can be checked against the Crime in India volume for the same year, while the framing paragraph is advocacy placed beside data, and it answers the member’s political question rather than the reader’s evidentiary one. The second is to notice what the question asked that the answer could not supply, because the gaps in the answers map the gaps in the statistical system. When a member asks how long the accused have been in jail and the ministry replies that the Bureau does not maintain such data, the exchange has produced a finding as significant as any figure in the annexure: the Indian state cannot say, from its own published statistics, how long the average person prosecuted under its principal anti-terror statute spends in custody. The third rule is to distrust every secondary report of an answer until the period and the table have been seen. News reports of these answers routinely carry the totals without the period, merge the figures of two answers given in different sessions, and convert the ministry’s person counts and case counts into a single undifferentiated mass. The errors catalogued in this article begin there far more often than they begin in the Bureau’s volumes.

The four denominators

The findable contribution of this article is a single table that fixes the vocabulary of the debate. Every figure quoted about UAPA enforcement is one of four measures, and each measure has a defined numerator, a defined source, and a characteristic error that arises when it is misused as the denominator of a rate.

Measure	What it counts	What it excludes	Where it is published	The error it invites
Cases registered	First information reports in which the UAPA is invoked, as returned by the states	Whether the UAPA charge survives to the charge sheet, and whether one case involves one person or twenty	Crime in India, and ministry answers that reproduce it	Treating cases as persons, which understates the human scale, since several accused are typically named in one case
Persons arrested	Individual accused persons taken into custody in cases where the UAPA is invoked	Persons named in the case who were never arrested, and arrests in which the UAPA is dropped before the charge sheet	Crime in India year tables; the ministry’s year-wise answers	Using it as the denominator for that year’s convictions, which compares two different cohorts and reports the system’s speed as its accuracy
Persons charge-sheeted	Accused persons against whom the investigating agency files a charge sheet, the step that commits the case to trial	Persons discharged before the charge sheet, and the time taken between arrest and charge sheet, which under the UAPA can run to 180 days	Crime in India; ministry answers for selected periods	Treating charge-sheeting as a measure of guilt, when it records only the agency’s decision to prosecute
Persons convicted	Accused persons convicted of a UAPA offence by a trial court	Acquittals, discharges, cases quashed, cases abated, and the far larger number whose trials are still pending	Crime in India disposal tables; ministry answers	Reading it as the year’s success count without asking how many of the year’s arrests had even reached verdict, which almost none had

Each row of that table deserves a paragraph of explanation, because the definitions contain the whole argument. Cases registered is the narrowest measure of the system’s activity and the broadest measure of its imprecision. A case is registered when a first information report names the Act, and the Act can be added to a report at registration and dropped before the charge sheet without any published table recording the change. The number of cases therefore tells the reader something about police behaviour at the moment of registration and almost nothing about what the courts eventually did. Persons arrested is the measure with the most human weight in it, because each unit is a person taken into custody, and the figure is the one that appears in headlines. It is also the measure most distant from the final outcome, because an arrested person’s case may take years to reach a charge sheet, more years to reach a verdict, and may end in discharge, quashing or acquittal. To place that year’s convictions beside that year’s arrests, as the standard public computation does, is to compare the output of trials that began years earlier with the input of arrests whose trials have mostly not begun.

Persons charge-sheeted is the measure the debate uses least and understands worst. The charge sheet is the investigating agency’s formal decision that there is material to prosecute, and under the UAPA it must be filed within the extended custody window the statute provides, which the complete guide to the UAPA describes in full. The charge-sheeted figure tells the reader how many of the arrested the agency thought worth prosecuting, and for the five-year period from 2018 to 2022 the ministry’s figures put that number at 6,503 out of 8,947 persons arrested. That is a genuine piece of information about the enforcement pipeline, and it is lost entirely in the standard presentation, which skips from arrest to conviction as though nothing happened in between. Persons convicted is the end of the pipeline, but the Bureau’s disposal tables count it alongside acquittals, discharges and quashings for the same calendar year, again mixing cohorts. The ministry’s answer to the December 2021 Rajya Sabha question set out the year-wise convicted figures next to persons in judicial custody at year end, which is the closest the official material comes to showing the reader the pipeline rather than its two ends.

One further distinction matters and is routinely lost, which is the difference between a case disposal and a person disposal. The Bureau’s tables count cases abated, quashed, discharged and acquitted as cases, while persons arrested and persons convicted are counted as persons. A ministry answer that prints both in a single table, as the December 2021 answer did, invites the reader to add and divide across categories that were never meant to be combined. The Bureau’s own conviction rate avoids this by conditioning on completed trials, and that is why the case-based rate of 27.57 per cent and the person-based quotient of 2.8 per cent can both be true of the same period without contradicting each other. They answer different questions. The honest reader asks the right one.

The table above also serves a second purpose, which is to give the reader a checklist for interrogating any quoted percentage. Ask first for the period, because the published answers cover different windows and the windows overlap. Ask second for the source, because a ministry answer and a Bureau volume are the same data at one remove, while a news report’s paraphrase is two removes and a social media post is three. Ask third for the denominator, because the four measures produce four different quotients and only one of them is a rate. Ask fourth for the cohort, because no published figure connects this year’s arrests to the eventual verdicts in those same cases. Any percentage that cannot answer those four questions is not evidence. It is atmosphere. The table and the checklist together are the article’s working equipment, and they are deliberately framed so that a reader on either side of the argument about the statute can use them without conceding anything in advance. The critic gains a version of the case that cannot be dismantled by a competent opponent. The defender gains a standard against which the system’s genuine achievements, in the cases that end in conviction after full trial, can be stated without inflation. The researcher gains a map of a statistical territory that is usually entered without one.

The five-year sets and what they actually say

The evidence base consists of a handful of published figure sets, each tied to a named answer and a stated period, and the first duty of a careful writer is never to merge them. Four sets cover most of the citable material, and they are set out here in the order in which a reader is likely to meet them.

The most authoritative set, because the Supreme Court worked from it, is the one the Minister of State in the Ministry of Home Affairs placed before the Lok Sabha on 2 December 2025, drawn from the Bureau’s volumes for the five years from 2019 to 2023. The year-wise arrest figures in that presentation are 1,948 persons arrested in 2019, 1,321 in 2020, 1,621 in 2021, 2,636 in 2022 and 2,914 in 2023, giving 10,440 persons arrested across the five years. The year-wise conviction figures in the same presentation are 34 persons convicted in 2019, 80 in 2020, 62 in 2021, 41 in 2022 and 118 in 2023, giving 335 persons convicted across the five years. Those two totals, 10,440 arrested and 335 convicted, are the figures behind the most widely repeated formulation of the record, and they should always be cited with the period and the source, because both the totals and their meaning change when either is removed.

The second set is the ministry’s answer in the Rajya Sabha covering the five years from 2018 to 2022, which is valuable precisely because it includes the step the other sets omit. Across those five years the answer recorded 8,947 persons arrested, 6,503 persons charge-sheeted and 252 persons convicted, and it recorded that only two cases under the Act were quashed by the courts across the entire period, both of them in Kerala in 2022. Set beside the first set, the overlaps are visible at once: the years 2019 to 2022 appear in both, the arrest totals differ because the windows differ, and any writer who adds the two arrest totals together has committed exactly the period-merging error that this article exists to prevent. The two sets agree in shape. Arrests run into the thousands each year, charge sheets follow at a level somewhat below arrests, and convictions in a year remain in the tens.

The third set is the year-wise table in the ministry’s answer to Rajya Sabha Unstarred Question 352, answered on 1 December 2021, which covers the years from 2016 to 2020 and adds a custody column that no other set provides. Persons arrested across those years were 999 in 2016, 1,554 in 2017, 1,421 in 2018, 1,948 in 2019 and 1,321 in 2020. Persons convicted were 24, 39, 35, 34 and 80 respectively. The same table recorded the number of arrested persons in judicial custody at the end of each year, which stood at 849 at the end of 2016, 1,283 at the end of 2017, 1,178 at the end of 2018, 1,317 at the end of 2019 and 1,091 at the end of 2020. That custody figure is a stock, not a flow: it counts the people inside on one day of the year rather than the people who passed through the system, and it demonstrates that at every year end in that window more than a thousand arrested persons remained in judicial custody under the Act, which is the enforcement record translated into its most concrete term. The same answer recorded the mixed character of the Bureau’s categories, printing the numbers of cases abated, quashed, discharged and acquitted as case counts in the same table as the person counts of arrests and convictions, and it recorded that the Bureau maintains no data in answer to a question about how many accused had been in jail for more than a year.

A fourth set, from a ministry answer covering the years 2016 to 2020 in a different presentation, recorded 5,027 cases registered and 24,134 persons who underwent trial in that period, of whom 212 were convicted and 386 were acquitted. That set deserves attention because the trial figure counts persons whose cases were before the courts, a population that includes the accumulated undertrial population carried across years, and because the near-parity of convictions and acquittals among the small number of persons whose trials were actually decided in the window is exactly the kind of figure that the completed-trial rate of the Bureau would predict. A fifth and earlier formulation, covering 2016 to 2019, gave 5,922 persons arrested and 132 persons convicted, and the figure of 2.2 per cent was computed by dividing the second total by the first. That computation is the origin of the most commonly quoted percentage in the debate. It has a named source and a stated period, which makes it better than most of what circulates, but it remains a ratio of two flows from different cohorts, and honesty requires it to be described as such whenever it is quoted.

Why do official answers give different totals for overlapping periods?

Because the published presentations cover different periods, draw on Bureau volumes published at different times, and are reproduced at different removes. An answer covering 2016 to 2020 aggregates one five-year window; an answer covering 2018 to 2022 aggregates a window shifted by two years; an answer covering 2019 to 2023 aggregates another five-year window.

Each answer is internally consistent with the Bureau tables for its own period, and each produces different totals for years that appear in more than one window, because the Bureau’s volumes for later years sometimes carry revised state returns. The reader who lays two totals side by side and finds them unequal has not necessarily found an error. The reader has usually found two windows that overlap without coinciding, and the correct response is to report each set with its period and source rather than to average, merge or prefer the one that suits an argument already held.

There is one more feature of the year-wise numbers that a careful reader should notice, because it explains much of the advocacy on both sides. The arrest numbers do not move smoothly. They rise from 999 in 2016 to 1,948 in 2019, fall to 1,321 in 2020, and then climb through 1,621 in 2021 to 2,636 in 2022 and 2,914 in 2023. Each movement has a context that the bare table does not supply. The 2020 fall coincides with a year in which ordinary policing and ordinary court functioning contracted across the country, and arrest numbers in many categories fell at the same time. The climb after 2021 coincides with large organised operations, most visibly the nationwide action against the Popular Front of India, which was declared an unlawful association in September 2022, and which generated arrests in many states in a short period. None of that context appears in the ministry’s annexure, which is a table and not an explanation. The reader who treats a change in the arrest total as a change in the underlying incidence of terrorism has added an inference the data cannot carry. The arrest figure measures enforcement behaviour, which responds to politics, operations, resources and court orders as much as it responds to the existence of offences.

The population figures for the National Investigation Agency cut in the opposite direction and are worth holding beside the state figures. The People’s Union for Civil Liberties, using the Agency’s own published case list, recorded that the Agency had registered 456 cases over the period it studied, of which 357 carried UAPA charges. The Agency’s caseload is small beside the state police registers, which run to thousands of arrests in a single year, and that asymmetry is itself a finding about where the statute is enforced. The UAPA in the official record is overwhelmingly a state police statute, applied by state forces to state-level offences, and the figures from the central agency that dominates commentary describe only a sliver of its actual operation.

The quality of the underlying data

A methodological account owes the reader a statement of the weaknesses in the raw material itself, because the Bureau’s tables are not a transcript of events but the end product of a reporting chain with known failure modes. The chain begins at the police station, where the first information report is written and the decision to invoke the UAPA is taken. It proceeds through district and state headquarters, where returns are compiled in the Bureau’s formats, and it ends in the published volume, sometimes more than a year after the year it describes. Three weaknesses in that chain bear directly on how the figures may be used.

The first is classification at the point of registration. The Bureau counts what the states return, and the states return cases as registered, which means that a case in which the UAPA is invoked at registration and dropped at the charge sheet stage appears in the arrest figures and never reappears as a correction. There is no published table of dropped UAPA invocations, so the record systematically shows the statute’s invocation at its widest and its adjudication at its narrowest, with the narrowing invisible. The second weakness is revision between volumes. State returns are occasionally corrected in later editions, which is one reason the same year can carry slightly different totals in answers given in different sessions, and why totals must always be cited from a named presentation rather than treated as constants of nature. The third weakness is the mixture of units within single tables, already noted in the December 2021 answer, where disposals are counted in cases and arrests and convictions in persons. A case with twelve accused that ends in acquittal enters the disposal column once and the arrest column twelve times, and no amount of care in reading the percentages removes the distortion that this asymmetry builds into any quotient formed across the two kinds of unit.

A fourth weakness arises where a case changes hands. A prosecution begun by a state police force may be transferred to the National Investigation Agency, and a case investigated by the Agency may be returned to the state, and each transfer moves the case between statistical universes that are compiled separately: the Bureau’s state returns, the Agency’s case list, and the ministry’s presentations drawn from each. No published reconciliation follows transferred cases across the boundary, so a researcher who adds the Agency’s register to the state totals risks counting the same prosecution twice, and a researcher who compares an Agency outcome with a state outcome may be comparing a case at two stages of its own life. The People’s Union for Civil Liberties study handled the two bodies of material as complements rather than as a single sum, which is the correct treatment, and the reader who encounters a combined total anywhere should ask how transfers were handled before believing it. The same caution applies to cases in which the UAPA is one of several statutes charged, since the Bureau’s classification of a multi-statute case can place it under a different head in different tables, and the Act’s apparent population can grow or shrink at the boundary of a classification rule without a single arrest having occurred or been undone.

None of these weaknesses makes the data useless, and it is a mistake to treat them as doing so. The reporting chain is the same chain that produces the statistics for every other offence, the classification conventions are stable across years, and the direction of the distortions is mostly known: registration-stage counting inflates the apparent reach of the statute, and cohort mixing understates the eventual conviction share of any given year’s arrests while overstating nothing about the length of the wait. A reader who knows the failure modes can state findings at the level of confidence the material supports, which in this field means findings about orders of magnitude, directions and persistent patterns, and not findings about precise percentages. The findings in this article are pitched at that level deliberately.

A civil liberties study reads the same volumes

The People’s Union for Civil Liberties published the most substantial independent reading of the enforcement data in its 2022 study of the UAPA, built from the Bureau’s Crime in India volumes for the years from 2015 to 2020 and from the National Investigation Agency’s published case information. The study’s arithmetic is the best documented in the field and it deserves to be stated in full before it is criticised. Across those six years the study recorded 5,924 cases registered and 8,371 persons arrested under the Act, against 235 persons convicted. It then computed two rates from the Bureau’s data and reported both. On the Bureau’s own convention, which conditions the rate on completed trials, it found a conviction rate of 27.57 per cent for UAPA cases against 49.67 per cent for cases under the Indian Penal Code. On the person-based computation, which divides persons convicted by persons arrested, it found 2.8 per cent against 22.19 per cent for the Penal Code. The study itself recorded the difficulty of correlating the cases registered in a year with the cases in which trial concluded in that year, and the parallel difficulty for persons arrested and persons convicted, which is the caveat on which the whole debate turns.

The study’s year-wise tables carry a further lesson that its headlines omit, which is the volatility of the completed-trial rate when the underlying numbers are small. On the case basis, the study’s computed rate moves from 14.5 per cent in 2015 to 33.3 per cent in 2016, to 49.3 per cent in 2017, back to 27.2 per cent in 2018, 29.2 per cent in 2019 and 21.1 per cent in 2020. Those swings are not changes in the character of prosecutions from one year to the next. They are the arithmetic of small denominators: in a system completing only a few dozen UAPA trials in a year, a handful of verdicts in either direction moves the annual rate by ten points. The same small-numbers warning applies with even greater force to single-state and single-year percentages, including the Jammu and Kashmir figures the Supreme Court quoted, where a rate below one per cent and a rate of three per cent can be separated by two verdicts. Aggregation across a five-year window is not a presentational preference. It is the minimum condition for the figures to mean anything at all, and it is one more reason the single-year percentage that circulates in the debate deserves no credence in either direction.

The comparison with the Penal Code is the study’s most quoted and least understood finding, and it cuts in two directions. On the completed-trial basis, a UAPA prosecution that reaches a verdict ends in conviction in somewhat more than a quarter of cases, while a Penal Code prosecution that reaches a verdict ends in conviction in roughly half of cases. That is a real difference and it is consistent with several explanations, including the difficulty of proving conspiracy and membership offences, the dependence on witnesses who turn hostile or cannot be protected, and the possibility that a substantial share of UAPA prosecutions were weakly founded from the start. On the person-flow basis, the contrast is starker, 2.8 per cent against 22.19 per cent, but that contrast partly reflects the far longer time UAPA trials take, because a slower system produces fewer completions of any kind in each calendar year and therefore fewer convictions to set against the year’s arrests. The study’s numbers are drawn from the same volumes as the ministry’s answers, and the organisation’s interpretation of them is its own. A reader who quotes the 2.8 per cent figure should attribute it to the study, state its period, and say what it divides by what, because the study’s authors did all three things and those who quote them usually do none.

The study also demonstrates, usefully and perhaps inadvertently, how easily the rhetoric on the critical side outruns the arithmetic. The inference drawn in much of the coverage of the study was that about 97.2 per cent of the 8,371 persons arrested, some 8,136 persons, had been acquitted. That does not follow from the study’s own data. A person who was not convicted in the window may have been acquitted, or discharged, or may still be awaiting trial, and for a statute whose trials run for years the third category is by far the largest. The study’s own tables show only a few hundred acquitted persons in the window, beside a few hundred convicted, with the great majority of arrested persons neither convicted nor acquitted but still inside the process. The corrected statement is less dramatic than the headline and more damning in a different way: the dominant experience of the arrested person under the UAPA is not acquittal after trial but indefinite suspension within a process that has not reached a verdict at all. Methodological honesty is not a favour done to the other side. In this instance it produces the stronger criticism.

Readers and researchers who want to lay these tables out side by side for a briefing can assemble the year-wise figures from the annexures and review the assembled tables in a presentation with the ReportMedic PowerPoint viewer, which is the practical way to check that the periods printed on each set have not been silently merged in the course of preparing the slides.

One year read three ways: the 2020 figures as a worked example

The method of this article can be compressed into a single exercise: take one year for which several presentations overlap, and read it the way a critic reads it, the way a defender reads it, and the way an analyst must. The year 2020 serves, because it appears in the December 2021 answer’s year-wise table, in the state-wise annexure for 2018 to 2020, in the People’s Union for Civil Liberties study, and in the ministry’s later five-year presentation, so every figure used here can be checked in at least two of those places.

The raw material for 2020 is as follows, each element with its unit attached. Persons arrested in the year: 1,321. Persons convicted in the year: 80. Cases acquitted in the year: 99, a case count, not a person count. Cases discharged in the year: 2. Cases abated in the year: 14. Persons in judicial custody at the end of the year: 1,091. The state annexure adds the distribution: Uttar Pradesh arrested 361 persons and convicted 54, Jammu and Kashmir arrested 346 and convicted 2, Manipur arrested 225 and convicted none, and Tamil Nadu arrested 92 and convicted 21, the last figure being the only substantial state conviction count in the annexure for that year outside Uttar Pradesh and Jharkhand.

The critic’s reading takes the first two numbers and divides: 80 convictions from 1,321 arrests is a conviction rate of about six per cent, and the year is pronounced a scandal of wrongful arrest. The defender’s reading takes the same two numbers and objects that the 80 convicted persons were arrested in earlier years, that the trials of the 1,321 have barely begun, that 2020 was a year in which the courts functioned under pandemic restrictions for months and every disposal figure in the country collapsed, and that the quotient therefore measures a disrupted calendar and nothing else. Each reading contains its characteristic error. The critic’s error is the familiar one of treating a flow ratio as a rate, compounded by ignoring that 2020’s conviction figure, at 80, is actually the highest of the five years in the December 2021 table, more than double the 34 of 2019. The defender’s error is to deploy the cohort objection as though it dissolved the whole table, when the table’s most important entry is the one neither side quotes: the 1,091 persons in judicial custody at the year’s end, a figure unaffected by the cohort objection, by the pandemic’s effect on disposals, or by any dispute about rates, which records that on the last night of 2020 more than a thousand people were in prison under the Act awaiting the conclusion of proceedings that, on the system’s demonstrated pace, would take years more.

The analyst’s reading holds all six numbers at once and states only what their combination supports. The system arrested 1,321 persons in a year in which its courts completed very few trials, convicted 80 persons and acquitted 99 cases, discharged almost nothing and quashed nothing, and carried a custody stock of 1,091 into the next year. The flows say the trial stage is the bottleneck. The stock says the bottleneck has a human population. The state figures say the population is not nationally uniform: two jurisdictions, Uttar Pradesh and Tamil Nadu, produced most of the year’s convictions, while two others produced hundreds of arrests and almost no convictions between them. Nothing in that reading requires a percentage at all, and it is more damaging to complacency, and more useful to reform, than either side’s headline. That is the standard to which every year in the published series can be held, and the reader who performs this exercise on two or three years will know the enforcement record better than almost anyone who writes about it.

The career of a percentage

The discourse angle of this subject deserves a section of its own, because the most quoted number in the debate has a traceable biography, and tracing it teaches the reader more than any general warning about statistics. The figure is 2.2 per cent. Its origin is a ministry answer covering the years 2016 to 2019, which gave 5,922 persons arrested and 132 persons convicted. Dividing the second number by the first yields 2.23 per cent, and that quotient, rounded and detached, became the conviction rate of the UAPA in news reports, in parliamentary speeches, in court submissions and in an uncountable number of social media posts. At each step of its travels the figure shed a qualifier. The period went first, so that a statement about four particular years read as a statement about the statute across its life. The source went next, so that a quotient computed by journalists from a ministry annexure acquired the authority of an official rate. The denominator went last and most completely, so that a ratio of two unrelated annual flows hardened into the proportion of prosecutions that succeed, which is the one thing it never measured. By the time the figure reached the Supreme Court’s discourse it had competitors: the People’s Union for Civil Liberties figure of 2.8 per cent for 2015 to 2020, computed the same way on a different window, and the Court’s own summary of 2 to 6 per cent for 2019 to 2023, computed the same way on the ministry’s presentation. Three percentages, three periods, one method, and a public debate that treats them as successive measurements of a single quantity.

The counter-claim has a biography too. The response that the figure is meaningless because of the time lag began as a correct methodological objection and hardened, in repetition, into a dismissal of the entire evidentiary field. In its strong form it asserts that since convictions in a year arise from arrests of earlier years, no relationship between the published totals can be asserted at all, and the record shows nothing. That overcorrects, and the overcorrection can be demonstrated from the objectors’ own premises. The lag objection explains why a single year’s quotient is unstable. It does not explain why, on the completed-trial basis that removes the lag entirely, the UAPA convicts at roughly half the Penal Code rate across a six-year window. It does not explain why the custody stock at each year end exceeds a thousand persons. It does not engage with the charge-sheeting proportion or with the geographic variance. A methodological objection that is valid against one computation has been stretched into an immunity against every computation, and the stretch is as much an artefact of advocacy as the original 2.2 per cent. The honest position is narrower than both slogans and stronger than either: the ratio is not a rate, the gap is real, and the gap’s size on any defensible computation is large enough to require an answer.

The pipeline between arrest and verdict

Between the two ends of the pipeline that the headlines quote, arrest and conviction, lies a sequence of stages at which a UAPA prosecution can end, stall or lose its UAPA character, and the published figures illuminate the stages unevenly. Mapping the figures onto the stages is the most useful exercise a reader can perform, because it shows exactly where the record goes dark.

The first stage is arrest and remand. The Act’s procedure provisions extend the ordinary timetable: investigation may continue with the accused in custody for up to 90 days in the first instance and up to 180 days where the court is satisfied on the public prosecutor’s report that the investigation requires the longer period, and the accused who is not charge-sheeted within the permitted period acquires a right to default bail. The arrest figures and the year-end custody stock belong to this stage. What no table records is how many of the arrested obtained default bail because the charge sheet was late, how many obtained regular bail, and how many remained inside, so the single most consequential sorting in the entire system, between those who wait inside and those who wait outside, is statistically invisible.

The second stage is the sanction and the charge sheet. A prosecution for the principal offences under the Act requires the prior sanction of the Central Government or the state government, as the case may be, granted on the recommendation of an authority constituted to review the evidence gathered, and the investigating agency’s charge sheet then commits the accused to trial before the special court. The charge-sheeted figure belongs to this stage, and in the 2018 to 2022 presentation it stood at 6,503 against 8,947 arrests. Sanction is a genuine filter in principle and a litigated one in practice: prosecutions have been discharged or stalled over defective sanction, most prominently in the proceedings against G. N. Saibaba, where the Bombay High Court discharged the accused in 2022 on the ground that the sanction for prosecution was invalid, an order the Supreme Court stayed within a day, after which a differently constituted bench of the High Court acquitted the accused on the merits in 2024. No published table counts sanction refusals, so the reader cannot tell from the official record how often the filter operates. The gap between arrests and charge sheets absorbs sanction refusals, evidentiary failures and decisions to drop the Act together, indistinguishably.

The third stage is the pre-trial sorting by the courts, and here the December 2021 answer’s disposal columns give the only published glimpse. In case counts, the courts acquitted 22 cases in 2016, 33 in 2017, 68 in 2018, 64 in 2019 and 99 in 2020; they discharged 2 cases in 2017, 23 in 2018, 16 in 2019 and 2 in 2020; cases abated, which includes cases ending because an accused died, numbered 4 in 2017, 4 in 2018, 3 in 2019 and 14 in 2020; and across the five years from 2018 to 2022 only two UAPA cases were quashed. Read as a pattern rather than as precise quantities, those columns show that the courts’ exits before or at trial are dominated by acquittals at the end of full trials rather than by early discharge or quashing, which means that the system’s sorting happens late, after the custody has been served, and not early, when it could prevent the harm. A prosecution that is going to fail under this statute generally fails after years, in a judgment, and not in its first months, on an application.

The fourth stage is the trial itself, conducted before designated special courts in the states and the Agency’s special courts for its own cases, and it is the stage about which the published record says least. The Bureau does not publish the number of witnesses examined, the number of hearings, or the age profile of pending UAPA trials. What is known from individual case records is the scale of the undertakings: prosecutions with witness lists running into the hundreds, in which the arithmetic of examination alone guarantees years. The Supreme Court’s bail jurisprudence has taken judicial notice of exactly this feature, treating the realistic impossibility of an early conclusion as a fact about the system rather than a misfortune of the individual case. The pipeline map therefore ends in an uncomfortable symmetry. The stages the figures describe well, arrest and charge sheet, are the stages controlled by the executive. The stage that determines outcomes, the trial, is the stage the figures describe worst, and the stage at which the system’s obligations are owed to the accused.

Special courts and the capacity question

The trial stage deserves a closer look, because the structural reasons for delay under the UAPA are specific and they compound. Trials under the Act are conducted by designated courts: the Agency’s cases go to special courts constituted under the National Investigation Agency Act, and state cases go to courts of session designated for the purpose, which in most states means an existing sessions court carrying the UAPA docket on top of its ordinary work. The judge is therefore rarely a judge of terrorism cases alone. The docket is shared with the ordinary criminal work of a district judiciary that is itself carrying one of the heaviest pending caseloads in the world, and a UAPA trial competes for court time against every other sessions trial in the district. Against that background the features of the statute’s own design do the rest. The offences are wide and are typically charged in combination, conspiracy alongside membership alongside fundraising alongside the substantive act, so a single case presents several distinct bodies of proof. The witness lists reflect the breadth: the investigation of an alleged network produces witnesses in the hundreds, each of whom must be examined and cross-examined for the trial to move. The evidence is frequently documentary and electronic in ways that generate their own satellite disputes about seizure, custody and certification. And the accused are often numerous and separately represented, so that the absence of one counsel, one accused or one witness on a date adjourns the whole.

None of these features is an excuse offered by this article. They are the mechanism by which the enforcement record is produced, and identifying the mechanism matters because it allocates responsibility correctly. A part of the delay is legislative: Parliament wrote offences whose proof is structurally slow and then restricted bail as though trials would be swift. A part is administrative: the state governments that invoke the Act have not built a special court infrastructure proportionate to their use of it, and the prosecution directorates that conduct these trials are shared with the ordinary system. A part is investigative: cases are charge-sheeted with witness lists and documentary records whose management the trial court then inherits. And a part is the ordinary condition of the Indian trial judiciary, which no statute has escaped. The comparative point is that the same legal system concludes Penal Code trials, on the completed-trial measure, at roughly twice the UAPA conviction rate and at volumes orders of magnitude larger, which indicates that the UAPA’s outcomes are not simply the judiciary’s outcomes. The statute’s design loads its trials with the heaviest proof burdens in the criminal calendar and then houses them in the ordinary machinery. The figures record the result.

The agency slice: two enforcement systems in one statute

One further figure set circulates in the debate and appears, at first sight, to contradict everything set out so far. The Ministry of Home Affairs, in a statement laid before the Rajya Sabha in March 2025 in answer to a starred question on the achievements of the National Investigation Agency, reported that the Agency had registered 652 cases since its inception, charge-sheeted 516 of them, arrested 4,232 accused persons and secured the conviction of 625, and that of 157 decided cases, 150 had resulted in conviction, giving the Agency a conviction rate of 95.54 per cent in its decided cases. The Agency’s decided-case rate has been reported at similar levels in other presentations, including a rate of 97.43 per cent in the 78 cases decided in the three years from 2022, given in the Lok Sabha in 2025. Set beside a national person-based quotient of 2.8 per cent, a figure of 95.54 per cent looks like a refutation. It is instead the best illustration in the entire field of why the denominator must always be named, because the two figures are computed on different populations, by different conventions, about different enforcement systems that happen to share a statute.

The Agency figure conditions on decided cases and counts cases, which is the Bureau’s own convention refined further by selection. The Agency chooses or is assigned a small number of cases, investigates them with concentrated resources, and its decided cases are those that have survived to judgment out of a register of hundreds, many of them still pending. An investigation of the Agency’s record published by The Wire added the composition that the headline rate conceals: of the Agency cases that had produced verdicts in the period it examined, a substantial share, 54 of 133, ended in guilty pleas rather than in contested trials, and pleas under a statute with the UAPA’s bail regime are entered by accused persons who have typically spent years in custody and for whom a plea may be the shortest available route out of prison. None of this makes the Agency’s rate false. It makes it a rate about a selected caseload resolved substantially by plea, which is a different fact from the proposition that UAPA prosecutions generally succeed. The state police systems that produce the national figures operate the same statute at the opposite scale, registering thousands of cases with ordinary investigative resources, and the distance between 95.54 per cent on decided Agency cases and the national picture is the distance between those two systems. A debate that cannot hold both numbers in view at once, each with its denominator attached, will use each in turn to shout down the other, and the record will be none the wiser.

The statistics that do not exist: bail, duration and the missing middle

A full account of the record must catalogue the figures the system does not produce, because the absences shape the debate as much as the tables do. There is no official count of bail under the UAPA: no table records how many accused applied, how many were refused, how many were released by trial courts, appellate courts or the Supreme Court, or how long each had spent inside before release. The bail jurisprudence of the statute, which fills volumes of law reports, has no statistical shadow at all. There is no official count of custody duration, as the ministry’s December 2021 answer conceded in terms. There is no official count of sanction decisions, of prosecutions dropped between arrest and charge sheet, of cases in which the UAPA was invoked at registration and abandoned before trial, or of the compensation, if any, paid to persons acquitted. There is no published age profile of the pending trial stock, so the reader cannot tell from official sources whether the average pending UAPA case is two years old or nine. And there is no linkage, at any level of the published system, between the person arrested in one year’s table and the person convicted or acquitted in another’s, which is the absence from which all the others derive their importance.

The consequence of the missing middle is that both sides of the debate are forced to argue from case anecdotes whenever they move from the aggregate to the human. The critic cites the student, the journalist or the professor whose prosecution collapsed after years, and the defender cites the convicted commander whose network the prosecution dismantled, and neither anecdote is answerable by the other because the population each is drawn from has never been enumerated. The enforcement record as published can support propositions about scale, direction and persistence. It cannot support propositions about typicality, and the discipline this article recommends is in large part a refusal to let aggregate figures pretend to a typicality they cannot establish, in either direction.

How the figures travel: the press as the third source

Between the ministry’s annexure and the public stands the news report, and the mutations the figures undergo in that passage deserve documentation, because for most readers the report is the record. The characteristic mutations are three. The first is the loss of the period, already described: a total for 2016 to 2019 becomes a property of the Act. The second is the loss of the unit, in which persons who underwent trial, a population swollen by the accumulated undertrial stock, are reported in headlines as though they were persons arrested in the period, so that the ministry’s figure of 24,134 persons who underwent trial between 2016 and 2020 travelled in reports as a trial population of which only 212 were convicted, inviting the reader to imagine 24,134 completed trials where the completed trials numbered in the hundreds. The third is the conversion of the quotient into a rate with an institutional voice attached: reports attribute the computed percentage to the government itself, writing that the government has admitted a conviction rate of two per cent, when the government has published two totals and admitted nothing of the kind, and has in other answers resisted precisely that characterisation. Each mutation increases the figure’s rhetorical force and decreases its evidentiary value, and each is avoidable by the elementary practice of citing the answer, the period and the denominator. The reader who applies that practice will find that most published versions of this record, on all sides, fail it.

The Supreme Court does the arithmetic in a bail order

The enforcement figures might have remained a quarrel among researchers had the Supreme Court not taken them up in a bail judgment. In Syed Iftikhar Andrabi v National Investigation Agency, a bench of Justice B. V. Nagarathna and Justice Ujjal Bhuyan granted bail to a man from Jammu and Kashmir who had spent more than five years in custody in a case investigated by the National Investigation Agency, in which the prosecution alleged a narco-terrorism conspiracy under the UAPA and the narcotics legislation. Bail had been refused by the special court and then by the High Court, substantially on the gravity of the allegations and the early stage of the trial. The Supreme Court disagreed, reaffirmed the holding of the three-judge bench in Union of India v K. A. Najeeb that Section 43D(5) of the UAPA cannot be the sole metric for denying bail where incarceration has become prolonged, and held that bail is the rule and jail the exception even under the anti-terror statute. What makes the judgment central to this article is that the Court supported that legal holding with the enforcement statistics, quoted from the same ministry presentation that this article has set out, the data placed before the Lok Sabha on 2 December 2025 drawn from the Bureau’s figures for 2019 to 2023.

The Court’s own summary of the numbers, as reported, was this: “Therefore, for all India figures, we have 2% to 6% conviction, meaning thereby that there is 94% to 98% possibility of acquittal in such cases in the country. In so far as the Union Territory of Jammu and Kashmir is concerned, the annual rate of conviction is always less than 1%. It means that at the end of the trial there is 99% possibility of acquittal in such cases.” The Court noted the components of that summary in the judgment: the all-India percentages computed from the ministry’s year figures run from about 1.5 per cent at the lowest to about 6 per cent at the highest across the five years, and the Jammu and Kashmir figures show a rate of zero in 2019 and a maximum of 0.89 per cent, recorded in 2022, with every year below one per cent. The Court then asked, in substance, whether continued detention was justified by the seriousness of the charge alone when the statistical prospect of conviction at the end of the process was of that order, and treated the statistics as one more good reason to follow Najeeb.

Two things about the Court’s use of the data deserve precise statement, because the judgment is cited in the debate as though the Court had certified the circulating percentage. The first is that the Court used the ministry’s own presentation, which computes each year’s percentage from that year’s arrests and that year’s convictions. The Court therefore used the ratio that this article has been at pains to distinguish from a true conviction rate, and it used it for a particular purpose: not to measure the accuracy of prosecutions across the system, but to assess the probability facing an individual undertrial whose trial had barely moved in five years. For that purpose the ratio is a defensible proxy, because an accused deciding nothing and a court deciding bail both care about the realised prospect that a prosecution of this kind ends in conviction, and the realised prospect is dominated by the system’s slowness as much as by its accuracy. The second point is that the Court’s reasoning does not depend on the percentage being exact. The legal work is done by the undisputed fact of prolonged custody without trial, and the statistics enter as corroboration that the delay is systemic rather than accidental. A critic who demonstrates that the true completed-trial rate is 27.57 per cent rather than 2 to 6 per cent has corrected the arithmetic without touching the judgment’s logic, because a system in which three quarters of completed trials end without conviction, and in which completion itself takes years, does not offer the undertrial a materially better prospect.

The Andrabi bench was building on a foundation that had itself been laid partly on the facts of delay. In Union of India v K. A. Najeeb, the three-judge bench that opened the constitutional route to bail under the Act was confronted with an accused who had been in custody for years while a trial citing witnesses on a very large scale proceeded at a pace that made an early verdict unrealistic, and the Court held that the rigours of Section 43D(5) melt where the Article 21 right to a speedy trial is infringed by incarceration of that order. The enforcement statistics entered that reasoning implicitly, as the background condition of a system in which such cases are routine. What the Andrabi judgment adds is the explicit step: the compiled record itself, quoted from the ministry’s own presentation, deployed as a reasoned part of the bail calculus rather than as atmosphere. A court that asks what prospect of conviction an undertrial faces, and answers from the published data, has made the enforcement record a legal fact, and every future bail contest in the statute will be argued in its shadow.

The judgment also matters for what it signals about the direction of the bail jurisprudence, and the reader who wants the doctrine rather than the data should follow the series article that owns it. The Court treated the statistics as a reason to insist that the Najeeb route remains open, resisted readings of the earlier decision in NIA v Zahoor Ahmad Shah Watali that would convert a prima facie assessment at the bail stage into something close to a finding, and expressed serious reservations about subsequent orders that had narrowed the Najeeb principle. The enforcement data and the bail doctrine meet at a single point: both are about what the state may fairly do to a person it has not yet proved anything against, over a period measured in years.

What the gap proves and what it does not

The argument over the figures is usually conducted as a choice between two inferences, and both are careless. The critical inference runs from the low quotient to the conclusion that the statute is misused on a mass scale. The defensive inference runs from the time lag to the conclusion that the figures are meaningless and nothing follows. The discipline of this article is to refuse both and to specify what survives.

The critical inference fails because a low conviction outcome is overdetermined. Convictions fail for reasons that have nothing to do with wrongful arrest: because organised violence cases depend on witnesses who are intimidated, killed or turn hostile; because conspiracy and membership offences require proof of agreements and states of mind that are genuinely hard to establish; because investigation quality in much of the state police system is poor; because trials are slow everywhere in the Indian system and delay degrades evidence in every category of case; and because some prosecutions are brought on material that was always thin. Only the last of those is misuse in any precise sense, and the aggregate figures cannot separate it from the others. The ministry’s standard answer makes a version of this point when it attributes conviction outcomes to the duration of trial, the appraisal of evidence and the examination of witnesses. That answer is self-serving and it is also partly true, and a serious critic has to concede the part that is true before pressing the part that is not answered.

The defensive inference fails for the opposite reason. The time lag explains why a ratio is not a rate, and it fully explains why no single year’s quotient should be quoted as one. It does not explain the size of the gap sustained across every published window. When the completed-trial basis is used, the UAPA still convicts in only about a quarter of the trials that finish, roughly half the completed-trial rate of the Penal Code over the same years. When the ministry’s own five-year totals are used, 10,440 arrests stand against 335 convictions, and even after every allowance for the trials still pending, the arithmetic requires that the overwhelming majority of those arrested will never be convicted of a UAPA offence, because the courts dispose of only a few hundred UAPA persons a year and the pending stock grows. When the custody stock is examined, more than a thousand arrested persons were in judicial custody at each year end from 2016 to 2020, in a system where the statute’s bail provision is designed to keep them there. The lag is a reason to correct the metric. It is not a reason to look away from the corrected picture.

What the data does support, stated as tightly as the evidence allows, is this. A very large number of people are drawn into custody under a statute that makes bail difficult to obtain, a meaningful proportion of those whose trials eventually finish are acquitted, and the interval between arrest and any verdict is measured in years rather than months for a substantial share of the accused. That proposition is the foundation on which the bail jurisprudence has been rebuilt, since it is exactly the situation of prolonged pre-trial detention under the UAPA that the bail interpretation article examines, and it is enough to carry a serious policy argument without inflating a ratio into a rate. The proposition also identifies where the burden of the statute actually falls. It falls before trial, not after it, and it falls on persons the state has not proved guilty of anything, which is why the enforcement record and the law of bail are two chapters of the same story.

One mechanism deserves to be named, because it connects the statistics to the statute’s design and because both sides of the debate have an interest in not naming it. Invoking the UAPA changes the accused’s position immediately and regardless of the eventual verdict: the custody window lengthens, the bail test hardens, and the practical prospect of release before trial contracts. The statute therefore delivers a large part of its practical effect, the incapacitation of the arrested person, at the moment of invocation, years before any court tests the evidence. An enforcement system working under such a statute faces a weakened institutional incentive to build cases to the standard a trial will demand, because the objective that is visible and immediate, taking a suspected person out of circulation, is achieved whether or not the proof is ever assembled. That is not an accusation against individual officers, and the figures cannot prove it about any individual case. It is a statement about incentives that the aggregate record is consistent with at every point: invocation is wide, charge-sheeting is broad, proof at trial is where the system fails, and the failure costs the system’s operators nothing like what it costs the acquitted. The enforcement record is what a rational observer would predict from the statute’s incentive structure, and that is precisely why the record, honestly read, is an argument about design and not only about administration.

There is a further point about charge-sheeting that complicates both standard arguments and that neither side quotes. In the 2018 to 2022 set, 6,503 of 8,947 arrested persons were charge-sheeted, which means that investigating agencies committed roughly seven in ten of the arrested to prosecution. A critic who portrays the system as arresting indiscriminately and releasing quietly has to account for that proportion, because the agencies did put most of their arrests before courts. A defender who portrays charge-sheeting as quality control has to account for what happened next, because the courts went on to convict only a small fraction of those committed. The pipeline loses most of its people at the trial stage, not at the arrest stage, and that location matters for reform: the failure the figures locate is a failure of proof at trial after long custody, not simply a failure of restraint at arrest, although the figures cannot acquit the arrest stage either, since roughly three in ten arrested persons were never charge-sheeted at all and had nonetheless been taken into custody under the Act.

The decided cases add texture to that analysis, because the judgments in acquittals and discharges recur to a recognisable set of failure modes, and the modes point in different directions. One mode is the failure of authorisation: sanctions granted without the review the statute contemplates, or granted by an authority whose competence is doubtful, which ended the Saibaba prosecution at the discharge stage in 2022 before the merits were reached on a later appeal. A second mode is the failure of the investigation’s physical record: seizures not proved, recoveries not connected to the accused, electronic material not authenticated, the defect the Supreme Court dwelt on when it acquitted the Akshardham accused in 2014. A third mode is the failure of the membership and conspiracy proof itself, where the prosecution establishes association, sympathy or presence and the court holds that the statute requires more, a gap between the breadth of the charge as framed and the content of the offence as construed that recurs across the case law. A fourth mode is simple attrition: witnesses resiling, witnesses dying, witnesses who cannot be traced after a decade, which is delay converting itself into acquittal. Only some of these modes describe misuse. The first two describe state failure in the conduct of prosecutions it chose to bring. The third describes a recurring mismatch between charging practice and judicial construction. The fourth describes time. A reform programme aimed at the record would have to work on all four, and a debate that treats every acquittal as proof of either persecution or incompetence will fix none of them.

The accumulation arithmetic deserves to be stated once, slowly, because it converts the whole debate into a single unavoidable observation. Take the ministry’s figures at their most favourable to the system’s defenders. Arrests in the published years run between roughly one thousand and nearly three thousand persons a year. Disposals of persons, convictions and acquittals together, run in the low hundreds a year on every presentation: 212 convicted and 386 acquitted across five years in one answer, a few hundred decided cases in the Bureau tables of the study period. Even if every pending trial ended tomorrow in the proportions the completed trials suggest, the courts would need many years of disposals at the demonstrated rate to clear the stock already accumulated, during which the arrests of further years would be added to the queue. The undertrial population under the Act is therefore not a backlog awaiting a one-time clearance. It is a standing population that the system’s inflow and outflow rates guarantee in equilibrium, renewed each year by the difference between what the police bring in and what the courts can finish. That is the deepest fact in the enforcement record, it requires no contested percentage to establish, and it is the fact that the Supreme Court’s bail jurisprudence, from Najeeb to Andrabi, has been steadily converting into doctrine.

What do the judicial custody figures in a ministry answer actually count?

They count a stock, not a flow: the number of arrested persons who were in judicial custody on the last day of the year named. The stock counts people inside on one day, not the people who passed through the system during the year, and the December 2021 answer gives the year-end figures across five years.

In that answer, the stock stood at 849 persons at the end of 2016, 1,283 at the end of 2017, 1,178 at the end of 2018, 1,317 at the end of 2019 and 1,091 at the end of 2020. A stock figure cannot say how many people entered custody during the year, how many were released, or how long any individual had been inside, and the ministry recorded in the same answer that the Bureau maintains no duration data at all. The stock figure is nevertheless the most humanly legible number in the official record, because it counts the people who were in prison under the Act on a given night, most of them awaiting a trial that had not concluded.

The geography of the record

The enforcement figures are not spread evenly across the country, and the concentration is one of the most cited and least carefully stated features of the record. The ministry’s state-wise annexure for the years 2018 to 2020, laid before the Rajya Sabha in August 2022, gives the clearest published picture. In that window the largest state total of arrests was recorded in Uttar Pradesh, with 479 persons arrested in 2018, 498 in 2019 and 361 in 2020, a total of 1,338, against 12, 17 and 54 convictions in those years, a total of 83. Manipur recorded 332, 386 and 225 arrests across the same years, a total of 943, with no convictions recorded in any of the three years. Assam recorded 331 arrests across the window with no convictions. Jharkhand recorded 368 arrests and 41 convictions. Among the union territories, Jammu and Kashmir recorded 177 arrests in 2018, 227 in 2019 and 346 in 2020, a total of 750, against two convictions across the whole window, both recorded in 2020. The later five-year set covering 2018 to 2022 identified Jammu and Kashmir and Uttar Pradesh as the states and territories with the highest numbers of arrests, and the Supreme Court’s judgment records that the Jammu and Kashmir annual percentage remained below one per cent throughout the 2019 to 2023 window.

A single enforcement event shows how these totals are made. In September 2022 the Central Government declared the Popular Front of India and its associated fronts unlawful associations, and coordinated operations by the Agency and the state police forces followed across a large number of states within days. Whatever the merits of that action, its statistical signature is unmistakable in the ministry’s own table: the national arrest total rose from 1,621 in 2021 to 2,636 in 2022 and 2,914 in 2023, the two highest figures in the published series. A ban-driven operation arrests on the membership and association offences, which are the offences whose proof at trial is the most contested in the entire Act, and it arrests in waves, at a scale set by the operation rather than by the maturation of individual investigations. The persons arrested in such a wave will reach their verdicts, if they reach them, across the following decade, and the conviction figures of 2028 or 2031 will be read by commentators who have forgotten the wave that produced the cohort. This is the cohort problem in its most concrete form, and it is the strongest practical reason for the publication reform proposed later in this article: without cohort tracking, every operational surge writes a statistical cheque that a later year’s commentators will try to cash as a verdict on the courts.

The concentration has to be read with the same discipline as the headline rate. It shows where the statute is used, and usage follows the geography of insurgency, of organised political violence and of the security operations mounted against both. Manipur and Assam carry long-running insurgencies in which the UAPA is a routine charging instrument. Jammu and Kashmir combines an insurgency with an intensive security presence. Uttar Pradesh has no comparable insurgency, and its position at the top of the arrest table in the 2018 to 2020 window reflects large-scale use of the Act in ordinary law and order and conspiracy prosecutions by the country’s largest state police force. The concentration therefore proves that the statute’s centre of gravity lies in a handful of jurisdictions, and that national averages conceal radically different local regimes. It does not by itself prove misuse in the high-use states, because high use is predicted both by genuine security conditions and by expansive charging practice, and the aggregate tables cannot separate the two. What the geographic figures do establish, beyond serious dispute, is that the outcomes differ as sharply as the inputs. A jurisdiction that arrests 943 persons across three years and records no conviction in any of them presents a question that its own enforcement authorities, and not only its critics, should have to answer, whatever the explanation turns out to be.

The variance between jurisdictions is itself a datum, and it is sharper than the variance in inputs. Jharkhand’s 41 convictions from 368 arrests across 2018 to 2020 and Manipur’s none from 943 are not two readings of one system; they are evidence that the charging thresholds, the quality of investigation, the composition of cases and the pace of the special courts differ so much between states that the national totals average away the phenomenon they purport to describe. A national conviction quotient for the UAPA is therefore doubly misleading: once because it is a ratio mislabelled as a rate, and again because the country it averages does not exist as a single enforcement jurisdiction. Anyone citing the national figure in an argument about a particular state’s record, or a state figure in an argument about the Act as a whole, has committed the geographic version of the period error.

There is also a federal point inside the geography that the national debate misses. Policing is a state subject, the UAPA is a central statute, and the Bureau’s tables are the record of state decisions to invoke central law. The field as a whole, from the lapsed statutes of the past to the standing Act, is traced in the series article on the lineage of India’s anti-terror laws, and the enforcement record adds the state dimension to that history: the same central text produces a near-dormant statute in most states and an intensive one in a few, and reform arguments aimed at the text alone will miss the level at which most of the record is made, which is the charging practice of a small number of state police forces and the special courts that serve them.

The predecessors in the same frame

The pattern in the figures is not new, and the People’s Union for Civil Liberties study placed the UAPA record in a line that the series’ history articles document statute by statute. The Terrorist and Disruptive Activities (Prevention) Act and the Prevention of Terrorism Act, both repealed after sustained controversy in which their enforcement records figured centrally, were likewise characterised by very large numbers of arrests against small numbers of convictions, and the conviction outcome of those statutes was a principal public argument for letting them lapse and for repealing them respectively. That history matters to a reading of the present figures in two ways. It shows that the Indian anti-terror enforcement system has produced the same statistical shape under three different statutory texts, which suggests that the shape is a property of the enforcement model, of broad definitions, restricted bail and slow special courts, rather than of any one Act’s drafting. It also shows what the political system does with such figures when a statute carries a sunset clause or a repeal campaign, and what it does not do when the statute is permanent. The UAPA has no sunset, its figures accumulate without a parliamentary moment at which they must be confronted, and the enforcement record is therefore debated in questions and answers rather than decided in a vote on continuance. The data’s audience has changed. Its shape has not.

After acquittal: the remedy that is not there

The enforcement record has a tail that the tables do not count at all, which is what happens to the person who comes out at the end of it. The Bureau’s vocabulary for exits is conviction, acquittal, discharge, quashing and abatement, and each of those words closes a file. None of them opens a claim. There is no statutory provision anywhere in Indian law that entitles a person who has been prosecuted under the UAPA, held in custody for a period measured in years, and then acquitted, to compensation, to an apology, to the restoration of employment or reputation, or even to a certificate that the prosecution failed. The absence has been put to the government in Parliament in terms, in questions that link the low conviction outcomes to the harassment of innocent persons, and the ministry’s replies have relied on the safeguards built into the statute and the constitutional remedies that already exist rather than on any scheme of redress for the acquitted. The policy gap is therefore not an oversight that a single amendment has missed. It is the settled position: the system prices the risk of a failed prosecution at zero for the state and at the whole cost for the person prosecuted.

What exists instead of a statutory scheme is a constitutional doctrine built case by case. In Rudul Sah v State of Bihar, the Supreme Court awarded compensation to a man who had remained in prison for years after his acquittal, treating the payment as a remedy for the violation of the right to life and personal liberty under Article 21 rather than as a private law claim. In Nilabati Behera v State of Orissa, the Court confirmed that compensation in public law for a constitutional violation is distinct from, and additional to, any remedy in private law, and that the defence of sovereign immunity has no place in it. Those decisions establish that a person wrongfully detained can petition the Supreme Court or a High Court and that compensation is an available remedy in an appropriate case. They do not establish an entitlement that operates automatically, a scale of payment, a limitation framework designed for the purpose, or any mechanism that reaches a person who has just walked out of prison after years inside and has neither counsel nor resources. Constitutional compensation for wrongful prosecution remains exceptional, discretionary and dependent on litigation that the acquitted are among the least equipped to mount.

The contrast inside the procedural law itself sharpens the point. The Code of Criminal Procedure obliges every state government to maintain a victim compensation scheme under Section 357A, under which victims of crime, including the dependants of those killed, can receive state-funded compensation on the recommendation of the courts or the legal services authorities, without any conviction being necessary. The Indian state has therefore accepted, in statutory form, the principle that public money should answer for some of the human costs the criminal process produces. It has accepted that principle for the victims of offences. It has not accepted it for the victims of its own failed prosecutions, although the second class’s loss is inflicted by the state directly and the first class’s loss is inflicted by a third party. That asymmetry is the cleanest statement of the policy gap in the entire record, and it requires no contested statistic to make.

The Law Commission of India addressed the gap directly in its Report Number 277 of 2018, on wrongful prosecution and miscarriage of justice, and recommended a statutory framework under which a person wrongfully prosecuted could claim compensation from the state through a designated court, with the claim extending to the pecuniary and non-pecuniary harm of the prosecution itself and not only of unlawful detention. The recommendation has not been enacted. The Akshardham temple attack prosecutions illustrate what the absence means in a concrete case. In Adambhai Sulemanbhai Ajmeri v State of Gujarat, decided in 2014, the Supreme Court acquitted all the accused who appealed, including men who had been sentenced to death and men who had spent the better part of a decade in custody, and criticised the quality of the investigation and the care with which the prosecution had been mounted. The acquittals corrected the convictions. Nothing in the legal system then addressed the years.

The undertrial side of the same gap is governed by a provision whose limits are rarely noticed. Section 436A of the Code of Criminal Procedure entitles an undertrial who has spent half of the maximum sentence for the offence in custody to release on bond, and district undertrial review committees exist to identify such prisoners. The provision carries an exclusion: it does not apply to offences for which death is one of the punishments specified, and the UAPA’s principal terrorist-act offence carries the death penalty where the act causes death. A large part of the UAPA undertrial population is therefore outside the one general statutory safety valve the system provides for excessive pre-trial detention, and its protection depends instead on the constitutional bail jurisprudence of Najeeb and its successors, which operates case by case and accused by accused. The enforcement record’s custody stock is thus not merely a failure of speed. It is the predictable output of a structure in which the statutory cap on undertrial detention excludes the statute’s own core offence and the remaining remedy is an individual constitutional claim.

Two practical routes exist for the person inside this gap, and both are partial. The National Human Rights Commission can inquire into a complaint of wrongful detention and recommend interim relief and compensation, which is one reason the remedial scheme of the Protection of Human Rights Act matters to this record, although the Commission’s recommendations are not binding and its jurisdiction over the armed forces is restricted. Free legal services through the legal services authorities are the route by which an undertrial obtains representation at all, and the framework of the Legal Services Authorities Act is what stands between most UAPA undertrials and total absence of counsel. Neither route compensates for a completed wrongful prosecution. The most concrete policy conclusion the enforcement evidence supports is therefore not about charging standards at all, although the evidence supports arguments there too. It is that a system which detains on this scale, convicts at this rate and takes this long owes the people it acquits a statutory answer, and at present it gives them none.

The comparative check: other special statutes in the same frame

Template discipline requires a comparative check, and the comparison that illuminates this record is not with foreign jurisdictions but with India’s other special enforcement regimes, which report their outcomes in the same two incompatible dialects. The Enforcement Directorate, which investigates money laundering under the Prevention of Money-laundering Act, has reported conviction outcomes in concluded trials at levels above ninety per cent, including 29 convictions in the 31 cases that completed trial between 2014 and 2023 on the Directorate’s own reported figures, out of more than seven thousand enforcement case information reports registered over its history. The National Investigation Agency’s decided-case rate of 95.54 per cent has already been set out. Set against those figures, the state police enforcement of the UAPA appears in the opposite dialect not because its officers are uniquely unsuccessful but because the central agencies report on the decided-case denominator, the denominator that flatters any prosecutor, while the national UAPA figures are debated on the arrest-flow denominator, the denominator that flatters any critic. The comparison disciplines both sides of the argument. It shows the defenders of the UAPA record that a high decided-case rate is what a selective, well-resourced prosecution service produces, and that the national system, which is neither selective in the same way nor resourced in the same way, cannot borrow the Agency’s number. It shows the critics that decided-case rates in the nineties are partly an artefact of guilty pleas and small decided populations, and that quoting the Directorate’s or the Agency’s rate as the standard the state system fails to meet repeats, in mirror image, the denominator error this article has been correcting throughout. The honest comparative finding is about structure, not virtue: boutique prosecution of a chosen caseload concludes with convictions; mass prosecution under a broad statute concludes, mostly, with waiting.

Oversight without a mechanism: what the record is for

The last structural fact about this record is the use to which the legal system puts it, which is almost none, and the contrast with the statute’s predecessors is instructive. The Terrorist and Disruptive Activities (Prevention) Act acquired, through the Supreme Court’s directions in Kartar Singh, a system of review committees charged with examining the application of the Act to pending cases, and the Prevention of Terrorism Act contained a statutory review committee whose examination of pending cases became a central event of that Act’s repeal, as the POTA article in this series records. Those mechanisms were imperfect, but they converted enforcement experience into institutional decisions: a committee that reviews cases can direct that the Act be withdrawn from them, and a Parliament that must renew a statute must confront its record. The UAPA contains no equivalent. It has no sunset clause requiring periodic reauthorisation, no review committee with power over pending cases, and no statutory obligation on any authority to publish or evaluate its enforcement outcomes. The entire public record described in this article exists because members of Parliament chose to ask questions and a civil liberties organisation chose to compile the answers. Oversight of the statute is therefore episodic and oppositional by construction: the figures surface when a member frames a question, and they subside when the session ends.

That institutional absence explains a feature of the debate that puzzles outside observers, which is its circularity. The same figures are produced, contested in the same terms, and laid aside, session after session, because no decision anywhere in the system is statutorily connected to them. The ministry’s formulaic answer, that safeguards exist and conviction is an elaborate process, is not merely a debating posture; it is the only response the structure calls for, since nothing follows institutionally from conceding or contesting the numbers. The Supreme Court’s intervention in the Andrabi judgment is significant partly because it broke the circle: a court took the record as a premise for a decision with consequences for an actual prisoner, and in doing so demonstrated the use the record can be put to when an institution is required to act on it. The bail courts are, at present, the only forum in which the enforcement figures have operational consequences, and they acquire those consequences one undertrial at a time.

The better measure and the honest verdict

If the conviction rate is the wrong measure, the obvious question is what the right one is, and the answer that the evidence itself suggests is time. The measure that would describe the enforcement of the UAPA faithfully is the distribution of time from arrest to disposal: how long the arrested person waits for a charge sheet, how long the charge-sheeted person waits for a verdict, and what fraction of the arrested population is still waiting at each anniversary of arrest. A system could defend a low conviction outcome if its trials were swift, because swift acquittal limits the harm of a failed prosecution. A system with slow trials and low convictions inflicts the punishment of imprisonment without the judgment that alone can justify it. Time to disposal converts the argument from a quarrel about a quotient into a question about a queue, and queues can be measured, published and shortened. The obstacle is that the published data does not support the measure. The Bureau does not track cohorts of arrested persons to their outcomes, does not publish the age of pending UAPA trials, and has told Parliament that it does not maintain the duration of custody. The single most useful reform in this field would cost no one a prosecution and frighten no one who believes the statute is well used: a published cohort table, in which each year’s arrests are followed to charge sheet, discharge, acquittal, conviction and continued pendency, year by year, until the cohort is exhausted. Until such a table exists, every participant in the debate is reasoning from the ends of a pipeline whose middle is invisible, and the methodological cautions in this article are the price of proceeding anyway.

The reform programme for the record itself can be specified in four publications, none of which requires legislation and none of which touches a single prosecution. The first is the cohort table already described, following each year’s arrested persons through charge sheet, discharge, acquittal, conviction and continued pendency until the cohort resolves. The second is a custody-duration distribution: for persons in judicial custody under the Act at each year end, the numbers who have been inside for under one year, one to three years, three to five years and more than five years, which the prison and court records from which the custody stock is compiled could support and which would convert the debate about prolonged detention from anecdote to measurement. The third is an accounting of the pre-trial filters: sanctions sought, granted and refused, and invocations of the Act dropped between registration and charge sheet, so that the gap between arrest and charge sheet becomes interpretable for the first time. The fourth is an age profile of pending trials, stating how many UAPA cases pending at year end were registered in each earlier year, which would let Parliament see whether the pending stock is a queue that moves or a sediment that accumulates. Each of these is a table the Bureau or the ministry could publish from records that already exist in some form, and the refusal or failure to publish them is itself part of the enforcement record, because a state that prosecutes under a statute of this severity owes the public the means to judge the prosecution. Until the tables exist, the honest analyst works with the ends of the pipeline and says so.

Is time to disposal a better measure than the conviction rate?

Yes, for the question that matters most, because it measures the burden the statute imposes on persons who have not been convicted of anything. A conviction rate, correctly computed on completed trials, describes the accuracy of finished prosecutions; time to disposal describes the wait that precedes any verdict.

That wait is the experience of the far larger group whose prosecutions never finish quickly: the years between arrest and verdict during which bail is restricted and custody is the default. The two measures answer different questions and a complete account needs both, but the harm documented across the enforcement record is overwhelmingly a harm of duration, and duration is what the conviction rate, on any denominator, cannot express.

For the working journalist or researcher, the discipline of this article reduces to a citation practice that can be applied in minutes. Quote no figure without its period, and print the period in the same sentence as the figure. Quote no figure without its source, distinguishing the Bureau’s volume, the ministry’s answer and the independent study, because the three carry different authority and different caveats. Name the denominator in words, persons arrested or cases completed, never allowing a percentage to appear whose base the reader cannot identify. Where a rate is genuinely meant, use the completed-trial rate and say that it is conditional on trials that finished. Where the flow totals are the point, print both totals and let the reader see the gap rather than a quotient. And where the human claim is the point, reach for the custody stock, which is the figure in the entire record that no methodological objection touches. A report that observes those rules can be as critical or as defensive as its evidence warrants, and it will survive correction, which is the only professional test that matters in a field where most of what is published cannot.

The verdict on the record, stated with the discipline the article has tried to practise, is as follows. The official figures, read with their periods and sources attached, establish that arrests under the UAPA run into the thousands each year, that convictions in any year number in the tens, that trials finish slowly and mostly without conviction, and that a standing population of arrested persons in custody at each year end is a permanent feature of the system rather than an aberration. The figures do not establish a precise rate of wrongful prosecution, and every attempt to extract one from them should be treated as advocacy rather than analysis. The figures do establish that the statute’s real operation is pre-trial: its severest effects are delivered by arrest, charge and delay under a restrictive bail regime, long before any court decides guilt. The reader who holds the argument at exactly that pitch, neither inflating the ratio into a rate nor dismissing the gap as an artefact, can use this data in any company without being corrected. That is the standard the implementation of a law should be judged by, and it is the standard this record, honestly read, meets the reader more than halfway toward a conclusion its defenders and its critics both have reasons to resist.

Frequently Asked Questions

Q: What is the conviction rate under India’s anti-terror laws?

There is no single official figure, because the answer depends on the denominator. On the National Crime Records Bureau’s own convention, which counts only trials that were completed, the People’s Union for Civil Liberties calculated a conviction rate of 27.57 per cent for UAPA cases from 2015 to 2020. On the computation that divides persons convicted in a year by persons arrested in that year, the same study found 2.8 per cent, and the Supreme Court, working from the ministry’s figures for 2019 to 2023, summarised the all-India position as 2 to 6 per cent. The first is a rate conditional on completed trials. The others are ratios of two unrelated annual flows, and should be quoted only with their period, source and denominator attached.

Q: Why is the UAPA conviction rate so low?

Several causes operate together, and the aggregate figures cannot separate them. Trials under the Act are slow, so few prosecutions complete in any year. The offences charged, conspiracy, membership and fundraising, require proof of agreements and states of mind that is intrinsically difficult. Cases depend heavily on witnesses who may turn hostile, be intimidated or become untraceable over the years a trial takes. Investigation quality across the state police forces that register most cases is uneven, and acquittal judgments recur to failures of sanction, of seizure proof and of electronic evidence authentication. Some prosecutions are also brought on material that was thin from the start. The ministry’s standard answer in Parliament attributes outcomes to trial duration, appraisal of evidence and examination of witnesses. A low rate is consistent with all of these causes in combination, and with some measure of over-charging, without proving any one of them.

Q: How long do anti-terror trials take in India?

No official average exists, because the National Crime Records Bureau does not publish time-to-disposal figures for UAPA cases and the Ministry of Home Affairs has told Parliament that the Bureau maintains no data on how long individual accused persons spend in jail. What the record shows instead is the shape of the delay. Persons in judicial custody under the Act numbered over a thousand at every year end from 2016 to 2020. In the bail litigation that produced the governing doctrine, the accused in Union of India v K. A. Najeeb had spent years in custody with a trial citing witnesses on a very large scale still far from conclusion, and the appellant in the Andrabi case had spent more than five years in custody with the trial at an early stage. Individual trials commonly run for years, and a substantial share of arrested persons wait longer for a verdict than the sentence they would have served on conviction for lesser charges in the same case.

Q: Is an arrest to conviction ratio the same as a conviction rate?

No, and the difference is the central methodological point about this record. A conviction rate, properly computed, is conditional on completed trials: of the prosecutions that reached a verdict in a period, the proportion that ended in conviction. On that basis the UAPA rate for 2015 to 2020 was calculated by the People’s Union for Civil Liberties at 27.57 per cent of completed cases. An arrest to conviction ratio divides the persons convicted in a calendar year by the persons arrested in that year, although the convicted were almost all arrested in earlier years and almost none of the arrested have yet been tried. That ratio measures the speed of the disposal system at least as much as the accuracy of its prosecutions. Both figures can be used honestly if they are named correctly, with the ratio described as a ratio and the rate reserved for the completed-trial computation.

Q: What happens to a person acquitted after years in custody?

In law, the person is released and the prosecution ends, and in practice almost nothing else follows. There is no statutory entitlement to compensation, an apology, restoration of employment or any certificate recording that the prosecution failed, however long the custody lasted. The acquitted person’s remedies are the exceptional ones built by the constitutional courts: a petition under Article 21 seeking compensation in public law, on the line of Rudul Sah v State of Bihar and Nilabati Behera v State of Orissa, which succeeds only in appropriate cases and requires fresh litigation by someone who has just emerged from years in prison. The Law Commission recommended a statutory compensation scheme for wrongful prosecution in its Report Number 277 of 2018, and it has not been enacted. The National Human Rights Commission can recommend interim relief on a complaint, but its recommendations are not binding.

Q: Which states register the most anti-terror cases?

The published state-wise figures place a small number of jurisdictions at the centre of the record. In the ministry’s annexure for 2018 to 2020, Uttar Pradesh recorded the largest state total of UAPA arrests, 1,338 persons across the three years, followed by Manipur with 943 and Jharkhand with 368, while among the union territories Jammu and Kashmir recorded 750. In the ministry’s presentation covering 2018 to 2022, Jammu and Kashmir and Uttar Pradesh were identified as recording the highest numbers of arrests. Most states record very few UAPA arrests in a year, and several record none, so the national totals describe the practice of a handful of police forces rather than a uniform national pattern. Arrests follow the geography of insurgency and of large-scale operations, and should always be read state by state rather than as a national average.

Q: Has the Supreme Court commented on anti-terror conviction rates?

Yes, in terms that gave the statistics a direct legal consequence. In Syed Iftikhar Andrabi v National Investigation Agency, a bench of Justice B. V. Nagarathna and Justice Ujjal Bhuyan granted bail to an accused who had spent more than five years in custody, quoting the data placed before the Lok Sabha on 2 December 2025 from the National Crime Records Bureau figures for 2019 to 2023. The Court observed that for all-India figures there was 2 to 6 per cent conviction, meaning a 94 to 98 per cent possibility of acquittal, and that in the Union Territory of Jammu and Kashmir the annual rate of conviction was always less than one per cent, meaning a 99 per cent possibility of acquittal at the end of trial. The Court treated those figures as a reason why the seriousness of the charge could not by itself justify continued detention, and reaffirmed the Najeeb holding on bail.

Q: Does a low conviction rate prove that a law is misused?

No. A low conviction outcome is consistent with misuse, but it is equally consistent with weak investigation, with the genuine difficulty of proving conspiracy and membership offences, with witness intimidation and attrition, and with the delays that degrade all Indian criminal trials. The aggregate figures cannot allocate the outcome among those causes, and a claim of misuse needs case-level evidence of the kind that acquittal judgments sometimes supply, such as findings of defective sanction or unproved seizures. What the figures do establish is a persistent and very large gap between the numbers drawn into custody and the numbers convicted, sustained across every published period and both major computations. That gap requires an institutional answer. It is not, without more, proof of any particular explanation, and treating it as proof weakens the criticism it is meant to support, because the methodological objection to the overclaim is sound.

Q: What is the difference between cases registered and persons arrested in the official UAPA figures?

A case is a first information report in which the Act is invoked, and a person arrested is an individual taken into custody in such a case, and the two counts diverge because one case typically names several accused. In the People’s Union for Civil Liberties study of the Bureau’s volumes for 2015 to 2020, 5,924 cases corresponded to 8,371 persons arrested. The case count measures the police decision to invoke the statute. The person count measures the human reach of that decision. Confusing them understates the number of people affected, and the confusion is built into some official tables, which print case counts for disposals, such as acquittals and discharges, in the same table as person counts for arrests and convictions, inviting readers to divide across two different units. Every figure in this field should be quoted with its unit stated.

Q: Does the NCRB track a UAPA case from arrest to its final outcome?

No. The National Crime Records Bureau compiles annual aggregates from state returns: cases registered and persons arrested in the year, and cases and persons disposed of in the year, without linking the two populations. A person arrested in 2019 who is acquitted in 2026 appears in the arrest column of one volume and the disposal column of another, and no published table connects the entries. That is why no genuine cohort conviction rate for the UAPA can be computed from published data, and why every published percentage is either conditional on completed trials or is a ratio of unrelated annual flows. A cohort table following each year’s arrests to their outcomes is the single most useful addition the Bureau could make to this record, and researchers have called for it without success so far.

Q: What does the persons charge-sheeted figure in a ministry answer actually count?

It counts the accused persons against whom the investigating agency has filed a charge sheet, the formal document that commits a case to trial before the special court. It is the middle term of the enforcement pipeline, between arrest and verdict, and it is the figure the public debate most often skips. In the ministry’s presentation for 2018 to 2022, 6,503 of the 8,947 persons arrested were charge-sheeted, which shows the agencies committing roughly seven in ten of their arrests to prosecution. The figure records the agency’s decision that there is material to prosecute, after the sanction the Act requires for its principal offences. It says nothing about guilt, and the later history of those charge-sheeted persons, among whom convictions number in the hundreds for the same period, is where the system’s losses are concentrated.

Q: What did the People’s Union for Civil Liberties find in its study of UAPA data from 2015 to 2020?

The organisation’s 2022 study, built from the Bureau’s Crime in India volumes and the National Investigation Agency’s published case list, recorded 5,924 cases registered and 8,371 persons arrested under the Act across the six years, against 235 persons convicted. It computed two conviction measures: 27.57 per cent on the Bureau’s case basis, which conditions on completed trials, against 49.67 per cent for Indian Penal Code cases, and 2.8 per cent on the person basis, dividing persons convicted by persons arrested, against 22.19 per cent for the Penal Code. The study itself noted the difficulty of correlating each year’s arrests with that year’s verdicts. Some coverage of the study inferred that about 97.2 per cent of those arrested had been acquitted, which the study’s own tables do not support, since most arrested persons had been neither convicted nor acquitted but were still awaiting trial.

Q: How do UAPA outcomes compare with conviction outcomes in ordinary Indian Penal Code cases?

On the like-for-like measures computed by the People’s Union for Civil Liberties from the Bureau’s volumes for 2015 to 2020, the UAPA underperforms the Penal Code on both. On the completed-trial basis, 27.57 per cent of UAPA cases that reached a verdict ended in conviction, against 49.67 per cent of Penal Code cases. On the person-flow basis, persons convicted in a year equalled 2.8 per cent of persons arrested in that year under the UAPA, against 22.19 per cent under the Penal Code. The second contrast is partly a measure of trial speed, since Penal Code trials complete far faster and therefore produce more verdicts per year. The first contrast is the more probative one, because it compares only finished trials, and it indicates that a finished UAPA prosecution succeeds at roughly half the rate of a finished ordinary prosecution over the same years.

Q: Why are UAPA arrests concentrated in a small number of states and union territories?

Because the statute’s use follows the distribution of the conflicts and operations it is used in, and the charging culture of a few police forces. States and territories with long-running insurgencies, Manipur, Assam and Jammu and Kashmir among them, use the Act as a routine charging instrument in security operations. Large operations produce arrest waves, as the nationwide action that followed the declaration of the Popular Front of India as an unlawful association in September 2022 did, and the national arrest total rose to 2,636 in 2022 and 2,914 in 2023 on the ministry’s figures. Uttar Pradesh, without a comparable insurgency, records high totals through volume of ordinary use by the country’s largest state police force. Concentration also means the national aggregates are misleading: outcomes differ so sharply by state, from Jharkhand’s 41 convictions from 368 arrests in 2018 to 2020 to Manipur’s none from 943, that there is no single national enforcement system for a national average to describe.

Q: Is there a statutory right to compensation after a wrongful UAPA prosecution?

No. No Indian statute entitles a person who has been prosecuted and acquitted, under the UAPA or generally, to compensation for the prosecution or the custody endured during it. What exists is the constitutional remedy developed by the Supreme Court, beginning with Rudul Sah v State of Bihar, under which compensation can be awarded in public law for a violation of Article 21 in an appropriate case, claimed by petition to a High Court or the Supreme Court. That remedy is discretionary, exceptional and dependent on fresh litigation. The procedural code does maintain state-funded victim compensation schemes under Section 357A for victims of crime, which makes the absence of any parallel scheme for the wrongfully prosecuted a visible asymmetry. The Law Commission’s Report Number 277 of 2018 recommended a statutory framework for wrongful prosecution claims, with a designated forum, and the recommendation remains unenacted.

Q: What is the difference between an acquittal, a discharge and a quashing in the enforcement figures?

The three are exits from a prosecution at different stages, and the Bureau counts them separately because they mean different things. An acquittal is a judgment after trial that the prosecution has not proved its case. A discharge is an order before trial, at the stage of considering the charge sheet, that the material does not disclose sufficient ground to proceed against that accused. A quashing is the termination of the proceedings by a High Court in its inherent or writ jurisdiction, usually on the ground that the prosecution is an abuse of process or discloses no offence. In the ministry’s presentation for 2018 to 2022 only two UAPA cases were quashed in five years, both in Kerala in 2022, and the December 2021 answer’s columns show discharges in single or double digits in most years, which together show that early judicial exits are rare in this system and that most failed prosecutions fail late, at judgment, after the custody has been served.

Q: Does any official source record how long UAPA accused spend in custody awaiting trial?

No published source does. When the question was put directly in the Rajya Sabha in December 2021, asking how many accused had been in jail for more than one year, the Ministry of Home Affairs replied that the National Crime Records Bureau does not maintain such data. The official record offers only a stock figure, the number of arrested persons in judicial custody at each year end, which stood at 849 at the end of 2016, 1,283 at the end of 2017, 1,178 at the end of 2018, 1,317 at the end of 2019 and 1,091 at the end of 2020. A stock cannot say how long any individual has been inside. Duration evidence exists only at case level, in bail orders recording that a particular accused has spent five or more years in custody, which is why the courts, rather than the statistics, have become the source of what is known about pre-trial delay under the Act.

Q: What would be needed to calculate a true conviction rate for the UAPA?

A cohort record: a table that follows a defined group of arrested persons from arrest to final outcome and reports what became of each of them. With such a table, the rate would be the proportion of one year’s arrested persons ultimately convicted, acquitted, discharged or still pending at successive anniversaries, and the long argument about denominators would end. The National Crime Records Bureau’s published aggregates cannot produce it, because they do not link a year’s arrests to later years’ disposals, and the Bureau would need case-level returns from the states with a persistent identifier for each accused person. Short of that, the honest alternatives are the completed-trial rate, which says how finished prosecutions end, and the custody and disposal flows, which say how the system is moving. Any single percentage offered as the true rate without cohort data should be asked for its denominator, its period and its source, in that order.

Q: How have annual UAPA arrest totals moved across the years reported to Parliament?

On the ministry’s presentations, arrests rose from 999 in 2016 to 1,554 in 2017, 1,421 in 2018 and a peak of 1,948 in 2019, fell to 1,321 in 2020, and then climbed through 1,621 in 2021 to 2,636 in 2022 and 2,914 in 2023, the highest figure in the published series. The 2020 fall coincided with a year in which policing and court functioning contracted across most categories of crime. The climb after 2021 coincided with large organised operations, most visibly the action following the declaration of the Popular Front of India as an unlawful association in September 2022. The totals measure enforcement behaviour, which responds to operations, resources and political direction as well as to the incidence of offences, so a rising arrest line is not by itself evidence of rising terrorism, and a falling one is not evidence of its decline. Each figure belongs to its named answer and period and should not be merged across presentations.

Q: Why does the National Investigation Agency report a conviction rate above ninety per cent while the national figures are so low?

Because the two numbers describe different populations on different denominators. The ministry told the Rajya Sabha in March 2025 that of 157 cases decided by the Agency’s courts, 150 had resulted in conviction, a decided-case rate of 95.54 per cent, from a register of 652 cases since the Agency’s inception, most of them not yet decided. The Agency selects or is assigned a small caseload, investigates with concentrated resources, and its rate conditions on the minority of cases that have reached judgment. An investigation of the Agency’s record published by The Wire found that of the cases that had produced verdicts in the period it examined, 54 of 133 ended in guilty pleas rather than contested trials. The national figures, by contrast, aggregate thousands of state police prosecutions, most still pending. Both numbers are real. Neither can be substituted for the other, and the distance between them measures the difference between a boutique prosecution service and a mass one.