₹13.97 crore moved without a sender record
In mid-October 2026, 187 payment instructions moved money out of Central Bank of India's own settlement account at the Reserve Bank of India, to 36 accounts across the banking system. The bank says its systems contain no record of ever sending them — and alleges entry through the central messaging server run by IFTAS, the RBI's wholly-owned technology subsidiary. IFTAS says no system of theirs was compromised. Both cannot be the whole story. We registered the incident, separated the corroborated from the merely claimed, and found a disclosure gap that does not depend on who is right.
- SFMS is India's domestic SWIFT. Since 2001, every inter-bank RTGS and NEFT payment instruction travels as a structured, PKI-signed message on it. The hub is run by IFTAS, a company wholly owned by the RBI (it took over from IDRBT in 2016).
- The money itself moves at the RBI. SFMS carries instructions; settlement happens in each bank's account at the RBI. Whoever can get a valid message accepted is the sender, economically — which is why message authenticity is everything.
- Three sanctioned doors (RTGS Regulations 4.21): the thick-client member interface over INFINET (the closed RBI-promoted network), a bank-gateway Web-API, and a browser PO module for small members — with public-internet access contemplated, so “INFINET-only” needs an asterisk.
- Every message carries an IDRBT-CA digital certificate, and the certificate rulebook says the key signing certificate belongs on a smart card or hardware token — by design, a stolen password alone should not move money.
- The gap: no public circular specifically mandates MFA for the SFMS console itself. Member-edge practices are enforced by audit, not by the hub — and how well that works varies by bank.
First, the three systems — a 90-second explainer
1. The settlement account. Every bank keeps a current account with the RBI. When customers of different banks pay each other over RTGS or NEFT, the banks settle among themselves through these accounts. It is the bank's own money — the working float of the payment system — not any customer's deposit.
2. SFMS. The Structured Financial Messaging System is the network that carries payment instructions between Indian banks — the domestic equivalent of SWIFT, in service since 2001, ISO 20022-based, carrying RTGS, NEFT, letters of credit and bank guarantees. Every bank runs its own SFMS server wired into its core banking system, and all of them connect to one central SFMS server.
3. IFTAS. The Indian Financial Technology and Allied Services — a Section 8 company incorporated 2015, which took over SFMS, INFINET and the banking community cloud from IDRBT on 1 April 2016, and has been wholly owned by the RBI since March 2019. The messaging backbone of Indian bank-to-bank payments is run by the central bank's own subsidiary.
The alleged fraud, in one sentence: messages appeared on this network saying "pay from Central Bank of India's settlement account", the money moved — and the bank's own systems show no record of ever having composed or sent those messages.
Doors and keys — what the rulebook actually requires
The complaint turns on one question: how hard is it, by design, to compose a payment message on a member connection? The governing documents are public, so we read them: the RTGS System Regulations v1.0 (Oct 2024), the certificate rulebook of IDRBT-CA — the RBI-promoted authority whose digital certificates sign every SFMS message — and the RBI circular that defines INFINET and SFMS membership as the entry ticket to payment systems. How that door is kept in practice at a member bank is a question for evidence, not rulebook — we put what one bank's real internal documentation shows in the Analysis tab. (That documentation comes from our separate Bank of Baroda leak verification; not the accused bank.)
Three documented doors into RTGS
Regulation 4.21 of the RTGS Regulations v1.0 defines how members reach the system. There are three access channels, not one:
- Thick-client / member-interface members connect over a dedicated, RBI-approved telecom medium — in practice INFINET, the closed financial network built by IDRBT and now run by IFTAS.
- Web / REST-API members connect through their own bank gateway — the core-banking-to-SFMS bridge.
- The "PO" module is browser-based, for members averaging 100 or fewer RTGS payments a day — and the rulebook contemplates public-internet access for it, under conditions the Bank is to prescribe.
So "INFINET-only" needs a qualification. All three channels ride INFINET or any other network the Bank approves; the closed financial network is the default, but the rulebook itself carries an internet door. What the rulebook does not publish is which member uses which door.
The keys: every message carries a certificate
SFMS is PKI-protected. Each message is signed with a digital certificate from IDRBT-CA (a licensed Certifying Authority since 2002). Its published Certification Practice Statement v3.2 (2012) defines Class 2 certificates for individuals and Class 3 for individuals and servers — the latter requiring physical appearance before a Registration Authority. For the System Certificate — the machine identity authenticating a bank gateway to the hub — §2.12.3 says the private key should be held in a secure token or smart card, with a hardware security module recommended for applications processing high-value transactions. In the designed model, sending a payment instruction requires a physical key object bound to a verified identity. A stolen password alone should not be enough — where that model is actually enforced.
What the rulebook does not say
- There is no public circular that specifically mandates multi-factor authentication for SFMS console logins — as of 26 Oct 2026 we found none. Annex 10 of the RTGS Regulations (Oct 2024) does require MFA for "users, applications, systems", a ban on internet-facing critical systems, and patch currency — but as generic participant duties, audited annually, not as product rules.
- There is no public register of which member bank uses which access channel, and no published MFA standard for the SFMS console itself.
- Certificate practice is set by a document from 2012 (CPS v3.2); what banks do inside their estates is their own affair until an audit says otherwise.
If you bank with Central Bank of India — or anywhere
- No customer lost a rupee. This was the bank's own settlement float, not deposits. Your account and your money are not what was attacked.
- Don't conflate this with the SBI story. SBI the same week confirmed ~3,897 customer records found on the dark web — a data-leak incident, a different failure class. Two state-owned banks, two different things, one confusing news cycle.
- Expect phishing riding the headlines. "Your RTGS failed", "your account is part of the bank fraud", "re-verify your KYC" — the bank will never ask for credentials or OTPs by call, SMS or link. Verify only through the official app or the number printed on your passbook. Report attempts to 1930 or cybercrime.gov.in.
- The systemic question is the regulator's to answer: if instructions can appear on the messaging network without a sender record anywhere, could the same be done from any other bank's server? SFMS carries lakhs of crores a day. That question deserves a public answer, not a two-party blame standoff.
The incident itself — what the police complaint says, our verification verdicts, the disclosure gap, one bank’s estate documentation, and method & sources — is in the Analysis tab.
- What happened: 15–17 Oct 2026, 187 payment instructions moved ₹13.97 crore out of Central Bank of India's own settlement account at the RBI into 36 accounts across 36 banks. The bank says none of these transactions exist in any of its own systems.
- Verified (L4): the transfers and the numbers — police complaint/FIR, five independent outlets, and CBI's own exchange filing reporting an outward-RTGS disruption in exactly that window.
- Disputed (L1): the breach vector. The bank claims entry via IFTAS's central SFMS server through the bank's own SFMS connection; IFTAS says no system of theirs was compromised. Nothing public settles it, and both sides hold the logs that would.
- Stands either way: the disclosure gap — a ~19 Oct “technical glitch” exchange filing, an FIR only on 22 Oct, against a 6-hour DAKSH incident-reporting clock — and 187 debits running three days on an account meant to reconcile daily.
- What one bank's estate looked like (BoB leak documentation — context, not accusation; not the accused bank): password-only SFMS logins until the Oct 2025 pen-test pass, a single admin password starting the sender/receiver processes, and 7 critical pen-test findings still open at report date.
- Customers lost nothing — this was the bank's own settlement float, not deposits. Would settle it: IFTAS hub logs + RTGS settlement audit trail for 15–17 Oct; a substantive CBI exchange clarification; a substantive IFTAS statement; the Maharashtra Cyber chargesheet.
What the police complaint says
On 22 October 2026, the bank's General Manager (Treasury) filed a complaint with Maharashtra Cyber; the Nodal Cyber Police Station, Mumbai, registered an FIR for cheating by personation. Per the complaint, as reported by Hindustan Times, Free Press Journal, Hellobanker, The Probe and The Wire:
- The outflow ran 15–17 October — 6 transfers on the 15th, 88 on the 16th, 93 on the 17th — totalling ₹13.97 crore to 36 beneficiary accounts, each at a different bank.
- The bank's forensic sweep found none of the 187 transactions in branch records, internet or mobile banking, the core banking system, its RTGS settlement records, or its own SFMS server.
- The only traces: partial entries for 10 transactions in SFMS database logs, whose transaction IDs match nothing in bank records. The 187 "sender" accounts named in the messages do not belong to the bank.
- Entry alleged: unknown persons reached IFTAS's central SFMS server through the bank's own SFMS server and generated the payment messages there.
- The bank notified RBI, IFTAS, the RBI's DAKSH incident-reporting portal, NCIIPC and CERT-In.
- A police officer told reporters that so little is known about how the transactions were generated that tracing the money is proving very difficult.
Our verification: two separate verdicts
The public is being handed one word — "hack" — for two claims of very different solidity. We scored them separately, using the same claim-level rubric as our other breach verifications (rumour → captured → corroborated → published).
| Claim | Asserted by | Our level | Why |
|---|---|---|---|
| The transfers happened: 187 / ₹13.97 cr / 36 accounts / 15–17 Oct | Bank complaint → FIR | L4 — corroborated | Police complaint + five independent outlets with consistent numbers + the exchange filing independently confirming a three-day outward-RTGS disruption in exactly that window |
| The vector: entry via IFTAS's central SFMS server, through the bank's own SFMS connection | Bank complaint | L1 — disputed | Asserted, not evidenced in public. IFTAS's reported position: "no compromise on any system." The hub logs that would settle it are held by the party denying |
The dispute is itself the story. If the bank is right, payment instructions can be injected into India's bank-to-bank messaging backbone from a member bank's connection without appearing in that bank's systems — a systemic finding about the payment network, not a ₹14-crore loss. If IFTAS is right, the compromise is inside one bank's own estate — and the bank is pointing upward. Neither party has published the one artifact that would settle the other's claim: the SFMS central-hub logs and the RTGS settlement audit trail for 15–17 October. Both sides hold the evidence that would exonerate them. Neither has shown it.
What we could not verify, and will not speculate on: the beneficiary accounts' status (frozen? who holds them?), any recovery, any cash-out geography or timeline (no captured source names any country — treat social-media versions as noise), and whether the bank met its reporting clocks (below).
The disclosure gap — the finding that stands either way
Whatever the forensics conclude, the paper trail of the bank's own disclosures is public, and it is uncomfortable:
| Date | What the public was told | Channel |
|---|---|---|
| ~19 Oct | "Technical glitch in RTGS outward system" — no cyber attribution, no amounts, no fraud | Stock-exchange filing |
| 22 Oct | Cheating by personation via alleged SFMS compromise; ₹13.97 cr; police FIR | Police complaint (reported publicly 24–25 Oct) |
Both statements cannot be equally complete. And they sit inside a regulatory frame that demands more: the RBI Master Direction on IT Governance (Nov 2023) and revised incident-reporting rules (effective Feb 2024) require banks to report cyber incidents to the RBI within six hours of detection on the DAKSH portal, plus proactive CERT-In notification — CERT-In's 2022 directions separately require six-hour notification. The bank says it reported to RBI, IFTAS, DAKSH, NCIIPC and CERT-In — but when it detected, and when it reported, it has not said. The rulebook's internal clock (hours) and the public's disclosure clock (days, euphemism first) are clearly different clocks. We flagged the same hygiene gap on the SBI leak story the same week.
The reconciliation question is just as sharp: after the 2016–18 SWIFT frauds (Bangladesh Bank $81M; Union Bank ~$171M attempted, caught at next-day reconciliation; City Union ~$2M; PNB ₹14,000 crore), the RBI mandated SWIFT–CBS integration and daily reconciliation. A settlement account is meant to reconcile daily. Here, 187 transfers ran across three days. Either reconciliation failed three days running — or bank and RBI records disagreed for three days without anyone noticing. The bank has not answered which.
Inside one bank's estate — what the “strong door” sits behind
The internal documentation we reviewed from the BoB leak — architecture manuals, operating SOPs, a 2025 penetration-test post-compliance report, and RBI-inspection evidence — shows what the "strong door" sits behind in practice at one member bank:
- A Windows application server running SFMS 7.0.1, with an Oracle 19c database behind it and message-queue links into core banking.
- Estate docs describe a single administrative password used to start the SFMS sender/receiver processes, and a manual "reprocess failed messages" path on the same console.
- The 2025 DFS-guideline penetration test across that NEFT/RTGS/SFMS estate: 284 findings, 28 critical — 7 critical still open at report date. The SFMS gateway application's own pen-test scored 3 High + 1 Medium, with 2 High open in Oct 2025 — including password-only logins for SFMS users, closed only during the Oct 2025 compliance pass. (The newer NEFT/RTGS console at the same bank did get OTP second-factor.)
- What does exist, and matters here: per-login access logs with source IP addresses at the bank site, reviewed monthly against an approved user-validation note — a replicable paper trail of who connected, from where, and when.
We publish this as context, not accusation. The accused bank is not Bank of Baroda, and nothing here says what happened on 15–17 October in CBI's estate. The point is narrower: the certificate-and-closed-network image of SFMS coexists, at member banks, with console-era practices that pen-tests keep flagging and compliance passes keep closing late. Whoever composed the 187 messages left a trail in one of two places — the member's own SFMS access logs, or the hub's. Both banks in this dispute hold such logs. That is exactly why the artifacts listed below would settle this.
What would settle this
- IFTAS hub logs + RTGS settlement audit trail for 15–17 Oct, disclosed or independently examined — one day of logs confirms or kills the vector claim.
- A substantive CBI exchange clarification replacing the "technical glitch" framing.
- An IFTAS statement with substance — what its logs show for the bank's connection on those three days, not a blanket denial.
- The Maharashtra Cyber chargesheet — 36 accounts across 36 banks is a mule-network signature; the chargesheet should map it.
We re-review on any of: a CBI clarification filing, a substantive IFTAS statement, the chargesheet, a recovery figure — or park the incident if 30 days pass in silence.
Method & sources
Read-only OSINT, 25 Oct 2026: press cross-check (five outlets), IFTAS corporate
history (IDRBT, RBI, Companies Act records), RBI Master Direction and CERT-In
reporting-requirement texts, exchange-filing coverage, precedent case records.
Extended 26 Oct 2026 with a documentation deep-dive for the doors
and keys section: RTGS System Regulations v1.0, IDRBT-CA Certification Practice
Statement v3.2, RBI access circulars and IFTAS service pages — plus internal technical
documentation from one state-owned bank running the same SFMS product family, reviewed
as part of our separate Bank of Baroda leak verification
(aggregate findings only; no individuals, no account data). We did not probe any bank
or IFTAS infrastructure and collected no beneficiary, customer or account-holder data.
Full claim review with evidence hashes:
CyberOpsec/reviews/2026-10-25-cbi-sfms-rtgs-heist-claim-review.md; incident
registered as cbisfms2026 on our breach-verification bench.
- Hindustan Times (24 Oct) — complaint numbers, IFTAS allegation, RTGS mechanics — hindustantimes.com
- Hellobanker (24 Oct) — fullest complaint reconstruction — hellobanker.in
- The Probe (25 Oct) — analysis, precedent frame — theprobe.in
- Free Press Journal (24 Oct) — no settlement-account entries — freepressjournal.in
- Economic Times — the "technical glitch" filing and the SBI disclosure context — economictimes.indiatimes.com
- CNBC Awaaz (Aseem Manchanda, 24–25 Oct) — IFTAS: no compromise on any system — x.com
- IFTAS corporate facts — idrbt.ac.in/iftas · Wikipedia: SFMS
- RBI six-hour DAKSH incident-reporting requirement — RBI Master Direction text
- RTGS System Regulations v1.0 (Oct 2024) — Reg 4.21 access channels, Annex 10 baseline cyber-security duties, Reg 6.10 settlement-account statements — RBI, canonical PDF
- IDRBT-CA Certification Practice Statement v3.2 — certificate classes (Class 2 individuals, Class 3 servers), §6.2.3 private keys on smart card/hardware token — cca.gov.in
- RBI, Access for Non-banks to Centralised Payment Systems (RBI/2021-22/73, 28 Jul 2021) — INFINET membership + SFMS use as the CPS entry ticket — rbi.org.in
- IFTAS SFMS service page — ISO 20022, INFINET as communication medium, LC/BG/E-FDR — iftas.in