Taiwanese city of Tainan will launch a pilot program in May 2025 that uses AI-driven facial recognition technology at select ATMs to address fraud.

Fraudsters often use money mules who hide their faces with masks or helmets, exploiting the lack of ID requirements at ATMs.

The new system will prompt users with a voice recording to remove masks or helmets for identification.

If identification fails within 10 to 15 seconds, an alarm will sound to alert bystanders or bank staff.

The system will not block transactions for unidentified users but aims to deter suspicious behavior through alerts.

🔗 Source: Focus Taiwan

🧠 Food for thought

1️⃣ Taiwan’s banking sector shaped by history of sophisticated ATM heists

Tainan’s new facial recognition initiative builds upon Taiwan’s hard-earned lessons from major ATM heists that exposed critical security vulnerabilities in its banking infrastructure.

In 2016, Taiwan experienced a sophisticated ATM heist where criminals stole $2.2 million by remotely commanding ATMs using malware installed through a compromised server in London 1.

This attack prompted Taiwanese banks to temporarily suspend withdrawals from approximately 1,000 ATMs of the same model, highlighting the systemic vulnerabilities in the system 1.

The criminals exploited the fact that approximately 90% of ATMs globally still operated on Windows XP at that time, an outdated operating system no longer receiving security updates 1.

Taiwanese police ultimately solved the case through a combination of surveillance footage, international cooperation, and a key tip from citizens who noticed suspicious activity at an ATM 2.

The shift to AI-powered facial recognition represents Taiwan’s latest countermeasure in an ongoing technological arms race against increasingly sophisticated financial criminals.

2️⃣ ATM security evolution mirrors broader cybersecurity patterns

Tainan’s facial recognition system represents the third major phase in ATM security evolution, reflecting how financial protection has progressed from physical to network to biometric defenses.