Cases of Banking Biometric Fraud Using Deepfakes Recorded in Kazakhstan

In Kazakhstan, fraudsters have repeatedly bypassed banks' biometric checks using deepfakes—fake videos created by AI. This threatens clients with loans taken out without their knowledge, and banks and microfinance organizations (MFOs) with liability for weak protection. The Agency for Regulation and Development of the Financial Market (ARRFR) explained these cases as violations of the law by the financial organizations themselves: the rules protect against AI fakes, but the reliability of the checks is the responsibility of the creditors.

How the verification works

According to ARRFR, bypasses have been recorded over the past three years, and the requirements apply to credit agreements in banks and MFOs. The system must randomly give the client at least three different commands and analyze their execution via video in real-time to filter out photos and pre-recorded videos. Authentication is considered failed after three unsuccessful attempts, and the match with the standard must be at least 95%. The program is required to monitor integrity, recognize emulators, and protect the camera from substitution.

Each procedure is accompanied by a document with results, the client's IIN, a list of commands, date and time; it is certified by the provider's electronic digital signature and stored in the financial organization. Biometric data is protected from copying and distribution, stored in encrypted form for at least five years. In case of a leak, ARRFR advises contacting the financial organization, filing a police report, and writing to the agency.

What changes in finance

Biometrics protect creditors from fake clients, and from 2027, large companies will undergo ESG checks before obtaining a loan. According to the Minister of AI and Digital Development, Zhaslan Madiev, the neural network will search for unjustified expenses in the budget, and four key financial systems will be integrated into a single ecosystem. The Smart Data Finance system already compares data from 80 sources and identifies discrepancies between people's income and their assets.