All notes
Fintech

Uzbekistan's central bank: 278.3bn soums lost to card fraud in nine months, malware APKs top the case count

Cardblock data for 1 January–1 October shows 113.1bn soums blocked. What the numbers say and what payment engineers should take from them.

What happened

On 13 November (page updated 16 November) the Central Bank of Uzbekistan published a quick analysis of fraud complaints registered on the Interior Ministry's Cardblock platform. It covers 1 January to 1 October 2026.

Headline figures:

  • Total losses: 278.3 billion soums (CBU).
  • Blocked through rapid action: 113.1 billion soums (CBU).
  • Number of cases: more than 39,000 (Spot.uz, Gazeta.uz).

Breakdown by method (CBU):

Method Cases Losses, bn soums
Malware installed on a mobile device 18,830 79.3
Following a scammer's instructions by phone 611 27.7
Telegram account takeover 2,808 21.9
Fake trading offers 1,044 15.1
Money sent to "financial pyramids" 1,086 14.0
Prepayment for goods/services never delivered 3,571 10.5
Online loans taken out in the victim's name 130 10.6
Payments in online games 272 0.9
Type not specified (debits from cards or e-wallets) 11,475 98.1

Separately, in 2,737 complaints (7.6 billion soums) people believed their money had been stolen, but it was actually an automatic loan repayment debit (CBU). Interior Ministry bodies are investigating the fraud cases.

Two caveats. Coverage of the period differs: the CBU says "1 January to 1 October", while some outlets wrote "10 months"; per the official source it is nine months. Also, the category losses add up to 278.1 billion soums versus the stated 278.3 billion — a small gap, probably rounding, which the CBU does not explain.

Context

The value of this data is that it shows where losses actually originate. By case count, malware leads — mostly APK files spread through messengers — at nearly half of all cases. By value, the largest line is "type not specified": the victim could not say how the money disappeared.

By my own arithmetic, the average loss varies a lot: roughly 4.2 million soums per malware case versus roughly 45 million soums per "instructions over the phone" case. Social engineering is rarer, but each case is far more expensive.

It is worth reading this next to figures released a few days earlier. According to Spot.uz, citing the CBU, banks' session anti-fraud systems flagged more than 3.86 million high-risk sessions in October alone, and 1.69 trillion soums of operations were restricted. That measures stopped attempts and restricted funds, not losses, so it cannot be added to or directly compared with the 278.3 billion. But it shows the defensive layer already operates at scale. The same article mentions a CBU draft from September: suspicious-transaction detection should run 24/7, score operations in real time and ask the customer for extra confirmation when in doubt. Public consultation has closed; there is no report that it has been adopted.

What it means

For users, the CBU's advice is concrete: don't install files from messengers or unknown sources, never share card details or confirmation codes with anyone — including someone claiming to be bank staff or police — turn on two-step verification in Telegram, distrust "multiply your money" offers, and contact your bank and the police immediately if defrauded.

For banks and payment companies, the main lesson is that an SMS code is no longer sufficient protection. Malware on the phone can read the SMS and confirm the payment on the customer's behalf. 18,830 cases is not a theoretical risk.

For engineers, a few practical directions stand out:

  • Device binding and attestation. First login from a new device, adding a card, raising a limit — these should be separate events requiring stronger confirmation. Play Integrity / App Attest, plus detection of overlay windows and Accessibility-based remote control, belong here.
  • Risk-based step-up. Not the same OTP for every payment: a new payee, an unusual amount and a recently changed password together should raise the risk score and trigger extra checks or a delayed transfer.
  • Architecture for fast blocking. Being able to block 113.1 billion soums shows that time is decisive. Propagating a complaint signal between banks and payment organisations within seconds, event-driven holds and idempotent block operations are all backend work.
  • Data quality. 11,475 cases and 98.1 billion soums under "type not specified" is the biggest line. Complaint forms and taxonomy need work: auto-attaching transaction ID, device, channel and payee would make the analysis far sharper.
  • Notification UX. 2,737 people took an automatic loan debit for fraud, which means the push/SMS text did not clearly say why money left the account. A small fix that reduces complaint volume and wasted investigative effort.

My take

Publishing this level of breakdown is the right move by the CBU: you cannot tune risk models without numbers like these. But in my view the next step goes beyond "citizens, be careful" towards stronger system-level controls. Malware and account takeovers start with a user mistake, yet the money still leaves through the bank's backend. That is where a payment can be stopped when a new device, a new payee and an unusual amount show up together.

My second point is about data structure. As long as the largest loss line is "unknown", it is hard to measure which defences work. Having the bank automatically attach technical context to a complaint would help both analysts and investigators.

If September's 24/7 real-time anti-fraud requirements are adopted, they will set a common floor across the market. The important thing is that they don't lead to excessive blocking of ordinary payments: good anti-fraud is invisible and fires only on real risk.

Sources: CBU (uz), CBU (ru), Spot.uz, Gazeta.uz, Spot.uz — anti-fraud, October. This is personal opinion, not investment advice.

// comments

Loading comments...

    // next step

    Let's build something together.

    Message me on Telegram
    aslbek@asldev: ~
    ❯