Most of This Year's Bank Fraud Did Not Happen This Year — And That Is the Whole Problem
Banking · Financial Crime · Policy · India 2026
Most of This Year’s Bank Fraud Did Not Happen This Year — And That Is the Whole Problem
Banks reported frauds worth ₹48,021 crore in 2025-26. Of that, ₹30,199 crore across 314 cases belongs to earlier financial years and was reported afresh after a Supreme Court ruling forced re-examination. The recurring question about Indian corporate failure is not who is next. It is why every stage of the response arrives late.
Every few years a large Indian corporate failure produces the same sequence of questions. Where was the regulator. Where was the board. Where were the auditors. Where were the banks. The names change, the sector changes, the questions do not, and that repetition is itself the finding worth examining.
The most useful way into it is not another list of cases. It is a single number in the Reserve Bank’s latest annual report, which quietly explains why accountability in Indian financial crime almost always arrives after the money has gone.
Quick Summary
Banks reported 10,114 fraud cases worth ₹48,021 crore in FY 2025-26, a 46% jump in value even as case numbers more than halved. ₹30,199 crore of that, or 63%, relates to earlier years. Meanwhile the average insolvency resolution now takes 744 days against a statutory 330, and creditors have realised roughly 32% of admitted claims under approved resolution plans. Detection, adjudication and recovery are each running years behind the event.
What We Know: the numbers on the official record
These are figures from the regulator’s own annual report and from replies placed before Parliament, not estimates or reconstructions.
- Fraud value rose sharply while case counts fell. Banks and financial institutions reported 10,114 cases involving ₹48,021 crore in 2025-26, against 23,722 cases involving ₹32,803 crore the previous year, a 46.4% rise in value on less than half the number of cases.
- A large part of it is historic. The RBI states that the 2025-26 figure includes fraud classification in 314 cases amounting to ₹30,199 crore pertaining to previous financial years, reported afresh after re-examination to comply with the Supreme Court judgment of 27 March 2023.
- The damage is concentrated in lending, not in apps. The advances category accounted for 8,640 cases worth ₹40,774 crore, roughly 85% of the reported value. Card, internet and digital payment frauds collapsed to 293 cases worth ₹29 crore, from 13,332 cases worth ₹517 crore a year earlier.
- Public sector banks carry the value. They reported 5,418 cases worth ₹35,709 crore, while private banks reported 3,956 cases worth ₹11,399 crore.
- The RBI is explicit about the lag. It notes that frauds reported in a year could have occurred several years earlier, and that the amount reported is not the same as the loss actually incurred, because recoveries reduce it over time.
Why the annual fraud number tells you almost nothing about the year
Look at reported value across five years and the line jumps around violently. It is tempting to read the peaks as crime waves and the troughs as improvement. Neither reading survives contact with how the number is actually built.
This is a measurement problem with policy consequences. A fraud committed in 2019, detected in 2022, contested through a Supreme Court ruling on borrower hearing rights, and finally classified in 2026 appears in the statistics seven years after the money left. Any system that reads those statistics to decide where to look is by construction looking backwards.
One number, three different meanings
The RBI publishes fraud data by date of reporting. Analysts usually want it by date of occurrence. Banks account for it by date of provisioning. The same rupee can be a 2019 event, a 2026 statistic and a 2024 write-off, which is precisely why public argument about whether fraud is rising or falling rarely converges. Before comparing any two fraud figures, check which clock each one is running on.
Where the ₹48,021 crore actually sits
That distribution matters because supervision follows attention. Digital fraud has produced visible institutional investment, including the RBI’s MuleHunter.ai model for identifying mule accounts in near real time and a cyber range platform for simulated drills. Credit fraud, which carries almost all of the value, depends on slower machinery: end-use monitoring, related-party disclosure, auditor scepticism and information sharing between lenders in a consortium.
The recovery arithmetic, stage by stage
Detection is only the first delay. What follows determines whether anything is actually recovered, and here the distance between the design of the law and its operation is measurable rather than rhetorical.
The consequence is not merely administrative. Value decays while a case sits. Rating agency analysis for FY 2025-26 found recoveries against admitted claims falling to roughly 23% from 46% the previous year, resolution plans approved dipping to 225 from 259, and liquidation recovery rates at about 4%. As of March 2026, nearly three in four pending insolvency cases had already run past the 330-day limit.
Two official recovery numbers that do not agree
During the passage of the insolvency amendment this year, the Finance Minister told Parliament that financial creditors have recovered more than 64% of their claims under the Code. IBBI’s own newsletters put realisation at roughly 32% of admitted claims. Both can be accurate, because they are measured against different denominators: one against the realisable or fair value of the assets that actually existed, the other against the full amount creditors claimed. The gap between those two denominators is the value that had already evaporated before the case was ever admitted, and citizens are entitled to be told which measure is being quoted.
Six places accountability leaks, and the evidence for each
- Detection lag. When 63% of a year’s reported fraud value belongs to earlier years, supervision is confirming history rather than interrupting it. Early-warning systems that trigger on repayment default trigger after the loss.
- Adjudicatory capacity. Resolution averaging 744 days against a 330-day statutory ceiling is a throughput problem before it is a legal one, and tribunal dockets extend well beyond insolvency into company law matters generally.
- Asset tracing across structures. Proceeds move through layered entities faster than attachment orders follow. The Enforcement Directorate reports proceeds of crime worth close to ₹65,000 crore attached across more than a thousand bank fraud cases under the money laundering law, with 150 arrests and 277 prosecution complaints filed.
- The distance between attachment and restitution. Attachment is not recovery. Assets worth ₹15,186 crore have been confiscated in fugitive offender cases, of which over ₹15,183 crore has been restituted to public sector banks. That is a genuine outcome, and it is also a fraction of the exposure in those accounts.
- Write-off without closure. A write-off is an accounting entry that does not extinguish the borrower’s liability, and the government has said so repeatedly. But ₹16.35 lakh crore removed from balance sheets over a decade, against ₹2.27 lakh crore recovered, means the pursuit continues long after the public conversation has ended.
- Deterrence timing. Fifteen individuals had been declared fugitive economic offenders as of October 2025, nine of them linked to large public sector bank frauds. Declarations of that kind come years after departure, by which point enforcement is negotiating with foreign jurisdictions rather than with a borrower.
| Stage | What the design assumes | What the data shows | The gap | Why it compounds |
|---|---|---|---|---|
| Detection | Fraud identified near occurrence | 63% of FY26 value is from earlier years | Years | Assets disperse before any order exists |
| Classification | Prompt reporting to the regulator | 314 cases reclassified after a court ruling | Multiple cycles | Statistics mislead the supervisor |
| Admission | 14 days to admit a case | Tribunal dockets run into thousands | Months | Operations deteriorate while waiting |
| Resolution | 330 days including litigation | 744 days average as on March 2026 | More than double | Realisable value falls with time |
| Realisation | Creditors made substantially whole | About 32% of admitted claims | Roughly two-thirds | Losses migrate to bank capital |
| Personal liability | Consequence follows wrongdoing | Declarations years after departure | Years | Deterrence weakens for the next case |
What has actually been built, and why it has not closed the loop
It would be inaccurate to describe the last decade as inaction. The insolvency framework created a time-bound process where none existed, and has resolved 1,376 companies with creditors recovering ₹4.11 lakh crore. The fugitive offender law created a mechanism for confiscating property when an accused leaves the jurisdiction. A separate audit regulator now exists. Enforcement activity under the money laundering law has expanded substantially.
What these additions have not yet done is compress the interval between the act and the consequence. Each new instrument attaches to a later stage of the sequence: the insolvency code operates once default is established, the fugitive law once someone has already gone, enforcement once proceeds have already moved. The stage with the least new machinery is the earliest one, when a loan is still performing on paper and the money is still traceable.
Follow the money means follow it early
Almost every recovery mechanism in the current system activates after value has left the perimeter. Real-time supervision, end-use verification of large credit, beneficial ownership transparency and consortium-level information sharing all operate before that point. They are less visible than an arrest and considerably less satisfying than a headline, and they are where the recoverable rupees actually are.
What Is Still Unclear
Four things cannot be stated with confidence from public data, and each of them limits how far anyone can reason about this.
- The true annual scale of fraud by year of occurrence. Published data is organised by date of reporting. Without an occurrence-based series, nobody outside the regulator can say whether the underlying incidence is rising or falling.
- How much of the reported amount is eventually lost. The RBI explicitly notes that the reported figure is not the loss, because recoveries reduce it over time. No consolidated published series tracks reported fraud through to final realisation.
- Which recovery denominator is the honest one. The 64% and 32% figures are both official and are not reconciled anywhere in the public record.
- Outcomes in individual cases. Investigations and prosecutions in most large matters remain pending before courts and tribunals. Amounts described as involved are allegations until adjudicated, and nothing in aggregate statistics establishes wrongdoing by any particular person.
Six changes that would compress the interval
Frequently asked questions
How much bank fraud was reported in India in 2025-26?
Banks and financial institutions reported 10,114 cases involving ₹48,021 crore, against 23,722 cases involving ₹32,803 crore in 2024-25. The number of cases more than halved while the value rose 46.4%. The data covers frauds of ₹1 lakh and above and is compiled by date of reporting, so a case appearing in this year’s figures may relate to an event several years earlier.
Why did the fraud amount rise when the number of cases fell?
Two things happened at once. Small-value digital payment frauds, which drove case counts in earlier years, collapsed from 13,332 cases to 293. At the same time large credit frauds dominated the value, and ₹30,199 crore across 314 cases relating to previous years was reported afresh after re-examination following the Supreme Court judgment of March 2023. Fewer, larger and older cases produce exactly this pattern.
How much do banks actually recover from corporate defaults?
Under approved insolvency resolution plans, creditors have realised roughly 32% of admitted claims according to IBBI data, with liquidation recovering far less, around 4% in the most recent year. The Finance Minister has separately told Parliament that financial creditors recover more than 64%, measured against the realisable value of assets rather than the full claim. Both figures are official; they answer different questions.
Why does insolvency resolution take so long in India?
The Code contemplates 330 days including litigation. The average for resolved cases was 744 days as on 31 March 2026, up from 713 a year earlier, and nearly three in four pending cases had already exceeded the limit. Published analyses cite tribunal capacity, litigation at multiple stages and the age of the underlying cases. Because realisable value falls as a business sits idle, delay is itself a cause of low recovery.
Does a loan write-off mean the borrower is let off?
No. The government has stated repeatedly in Parliament that a write-off is an accounting step taken once a bad loan has been fully provided for, that it does not waive the borrower’s liability, and that recovery action continues. Scheduled commercial banks wrote off about ₹16.35 lakh crore over a decade, and public sector banks have reported recovering ₹2.27 lakh crore from written-off accounts, which shows both that pursuit continues and how partial it is.
What is the Fugitive Economic Offenders Act and has it worked?
It allows a court to declare a person a fugitive economic offender where they have left India to avoid prosecution in a specified economic offence, and permits confiscation of their property including benami assets. As of October 2025, 15 individuals had been so declared, nine of them linked to large public sector bank frauds. Assets worth ₹15,186 crore have been confiscated and over ₹15,183 crore restituted to public sector banks, against a stated loss of ₹58,082 crore in those accounts.
Is digital payment fraud the biggest financial crime risk in India?
Not by value. In 2025-26 the card, internet and digital payments category accounted for 293 cases worth ₹29 crore, while the advances category accounted for 8,640 cases worth ₹40,774 crore, roughly 85% of the reported total. Digital fraud affects far more individuals and deserves the attention it receives, but the systemic loss sits in corporate lending.
Who ultimately bears the cost of a large corporate failure?
The loss moves rather than disappearing. It lands first on lender capital, which for public sector banks means the government as majority shareholder and therefore the taxpayer. It reaches shareholders through impaired valuations, employees through job losses, and suppliers as operational creditors, who historically recover a smaller share than financial creditors. The least measurable cost is the erosion of confidence, which raises the price of credit for borrowers who did nothing wrong.
The short version
The recurring question is usually framed as a search for the next offender. The data points somewhere less dramatic and considerably more fixable. Sixty-three per cent of this year’s reported bank fraud value belongs to earlier years. Resolution takes more than twice the time the statute allows. Creditors realise about a third of what they claim, and liquidation returns almost nothing. Each of those is a measurable interval between an act and its consequence, and every one of those intervals is where recoverable value quietly disappears. Strong institutions are not primarily about the severity of punishment. They are about how quickly the system notices, decides and collects, and those three numbers are published every year for anyone willing to read them.
