Part 1 | Andrabi’s unprecedented reliance on conviction data for granting bail under UAPA

This commentary is divided into two parts (Read Part 2 here). Part 2 examines Andrabi’s rejection of what has become a standard State offer: delaying the grant of bail by one more year.

On 18 May 2026, a Division Bench of Justices B.V. Nagarathna and Ujjal Bhuyan, in Syed Iftikar Andrabi v NIA (Andrabi) granted bail to an undertrial prisoner—Syed—who was incarcerated for over five years and eleven months on charges of narco-terrorism under the Unlawful Activities (Prevention) Act, 1967 (“UAPA”) and the Narcotic Drugs and Psychotropic Substances Act, 1985 (“NDPS Act”). In the process, the Bench criticised two rulings—Gulfisha Fatima v State (2026) (“Gulfisha”) and Gurwinder Singh v State of Punjab (2024) (“Gurwinder”) for hollowing out a three-judge bench decision, Union of India v. K.A. Najeeb (2021) (“Najeeb”). Najeeb held that statutory conditions of bail under the UAPA would “melt down” in light of prolonged incarceration. Gulfisha, which denied bail to activist Umar Khalid, and Gurwinder were authored by Justice Aravind Kumar.

Five days later, on 22 May, a different Division Bench of Justices Kumar and P.B. Varale in Tasleem Ahmed v State (NCT of Delhi) (“Tasleem”) referred the question of how Article 21 of the Constitution is to be applied against the statutory bar on bail under Section 43-D(5) of the UAPA to a larger bench. This was done at the request of the Union government, with the Additional Solicitor General S.V. Raju pointing to the Andrabi decision.

Both decisions flow from the same upstream fault line in the Court’s jurisprudence. In Andrabi, the Court criticised Gulfisha and Gurwinder. In Tasleem, the Court, rather than add a further competing formulation, referred to the conflict for “authoritative resolution.”

This commentary examines the unprecedented reliance on data in Andrabi as an argument against pre-trial detention in bail jurisprudence under special statutes.

The Jurisprudential Map

Before these issues are addressed, it is useful to map the landscape that produced the reference. The conflict is not simply about Najeeb, but about what Najeeb meant and how successive benches of equal strength managed, or failed to manage, their disagreement with it.

The table below summarises how successive benches of this Court in 27 judgements (using SCC Case Reference tool) have applied, or declined to apply, the principle in Najeeb from 2021 to 2026. The pattern is not random. 

It reveals three distinct lines of cases: 

1. Narrowed Najeeb,
2. Accepted Najeeb but rejected bail,
3. Applied Najeeb faithfully

It also attempts to illustrate that the reference in Tasleem addressed a pot that has been boiling for at least two years—the doctrinal conflict predates both Gulfisha and Gurwinder.

The conflict is not merely Gurwinder and Gulfisha on one side and Najeeb/Andrabi on the other. It runs through at least four years of Division Bench decisions, including Harpreet Singh Talwar and the split decision in Mohd. Tahir Hussain, which quietly narrowed Najeeb without the same degree of explicitness as Gurwinder and Gulfisha. The reference in Tasleem, in other words, addresses a structural problem in the Court’s institutional management of precedent, not merely the specific formulations of two particular benches.

What Andrabi holds and why it matters

In Andrabi, the Bench holds that a bench of lesser strength cannot dilute, circumvent, or disregard the ratio of a larger Bench. Both Gurwinder and Gulfisha had “made a clear departure from the ratio laid down in K.A. Najeeb” without expressly disagreeing with it. 

Gurwinder’s twin-prong test requires the Court to assess whether the accusation against an accused is prima facie true. If not, the Court can only then proceed to conventional bail considerations. Such a reading neither flows from the text of Section 43-D(5) nor from Najeeb. In fact, the Andrabi bench held that it is in direct conflict with Najeeb. Accepting the twin-prong test would result in pre-trial incarceration acquiring a post-trial punitive character. No matter how long the accused remains in custody, bail becomes permanently unavailable once the prima facie threshold is crossed. 

Andrabi also makes a point that deserves to be stated clearly: the principle that ‘bail is the rule and jail is the exception’ is not an empty slogan flowing from the Code of Criminal Procedure, as characterised by Gurwinder. The bench held “(i)t is a constitutional principle flowing from Articles 21 and 22 of the Constitution and the presumption of innocence.” Legislation may calibrate its application, but it cannot “invert the constitutional relationship between liberty and detention”. Section 43-D(5) operates as a restriction, not a reversal. 

This has a specific implication that Andrabi didn’t develop. The UAPA is not a preventive detention statute. It is a penal statute providing for prosecution, trial and punishment. An accused held under UAPA is an undertrial, not a detenu in the constitutional sense. Article 22(1) and (2) of the Constitution, which confer specific rights on arrested persons and which operate within the architecture of a penal prosecution, apply in full. The entitlement to bail flows from this architecture, and restrictions on it require corresponding constitutional justification, not merely parliamentary sanction.

Conviction data as a constitutional argument? 

Andrabi also did something that no other judgment concerning bail under a special statute has done before. It cites National Crime Records Bureaus figures for the years 2019 to 2023 as an independent and explicit basis for questioning the justifiability of continued pre-trial detention. The all-India picture shows that the conviction rate under the UAPA ranged from a low of 1.56% in 2022 to a high of 6.06% in 2020. The Jammu and Kashmir figures are more severe: in two of the five years, the conviction rate was zero. For the period as a whole, the annual conviction rate in the state never exceeded 1%. With the said statistics, the Bench frames a pertinent argument, “should we continue the detention of the appellant or defer the consideration to a later stage, simply because the charges are serious?”

 The constitutional architecture of the argument

The conventional framework for bail under Section 43-D(5) of UAPA requires the court to assess whether there are “reasonable grounds for believing that the accusation against such person is prima facie true.” This is an assessment of the prosecution’s case as presented, with the chargesheet and case diary taken at face value. The conviction data challenges not any individual application of this standard, but the standard’s systemic relationship to actual guilt.

The argument can be formulated precisely. The justification for Section 43-D(5)’s stringency rests on a premise that persons whose cases satisfy the prima facie threshold pose a genuine risk to society of the kind that justifies departing from the favour of liberty. However, this premise is not empirically supported if  94 and 99 percent of persons are ultimately acquitted. 

The Maneka Gandhi framework, that any procedure depriving a person of liberty must be fair, just, and reasonable, has historically been applied to the design of the procedure rather than to its empirical outcomes. Andrabi gestures towards applying it to outcomes.

What does the above argument not resolve?

In Andrabi, the Court does not take the data argument to its logical conclusion. It does not hold Section 43-D(5) disproportionate, and it does not develop a framework for when conviction rate data becomes relevant. Two limitations deserve acknowledgment. 

The first is methodological. The NCRB figures measure persons convicted against persons arrested in a given calendar year, not against persons whose trials concluded in that year. The numerators and denominators are drawn from different cohorts, which means the 1.56% figure for 2022 does not mean that 98.44% of trials that concluded in 2022 resulted in acquittal; it means that the number of convictions in 2022 was 1.56% of the number of arrests in 2022. The correct figure for acquittal rates would require cohort tracking across years, which the NCRB data does not provide. The Court does not flag this caveat. Moreover, the Court should have also noted the figures of cases registered, delays in filing a mere charge-sheet and number of acquittals, which is also available in the Crime in India Reports and Lok Sabha’s Q/A. 

The second limitation is one of relevance. Low conviction rates may reflect many factors besides the weakness of individual cases, including witness protection failures, prosecutorial resource constraints, delays in forensic analysis, and the technical difficulty of establishing terrorist conspiracy charges at trial. Not all of these factors imply that the prima facie material in individual cases is insufficient. Courts applying the data argument should be careful not to treat aggregate acquittal rates as establishing that any particular accused will be acquitted. These limitations do not, however, undermine the core methodological contribution. 

Andrabi is one of the first bail judgements under a special statute to treat aggregate prosecutorial outcome data as constitutionally relevant. The numbers, now read into the record by the Court itself, are available for development by future benches.