← Cases
TaiwanFinance and insurance2025 – ongoing

Far Eastern International Bank

AI-supported multi-layer fraud control for virtual asset flows

Far Eastern International Bank says its fraud controls blocked over NT$300 million in suspicious transactions

WorkingCompany's own claim

Problem

Far Eastern International Bank handles fiat flows for virtual asset exchanges in Taiwan. These flows carry fraud risk, and many potential victims either do not report a scam or do not believe they were deceived.

Approach

The bank built a multi-layer framework with warnings before transactions, interception during them and tracking afterwards. It says it adopted AI to analyse transaction behaviour, amounts and timing, together with external risk databases and cross-institution data. Exchanges send real-time alerts by API, and the bank works with the National Police Agency's Criminal Investigation Bureau. Branch staff can also stop suspicious transactions.

Results

more than NT$300 millionsuspicious transactions blocked by the bank's own fraud controls, as stated by the bankCompany's own claim[1]
more than NT$50 millionfraud funds returned to victims by October 2025, as stated by the bankCompany's own claim[1]
70%share of flagged suspicious accounts that became alert accounts within 90 days, as stated by the bankCompany's own claim[1]

Takeaway

The bank counts blocked and returned funds as the results of the whole framework, which combines AI analysis with police data sharing and branch staff. The page does not say how much AI alone contributed.

Categories

Business function
Finance, risk and legal
Innovation type
Process innovation
Evidence
Company's own claim

Sources

Source summaries

[1]TechNews8 Jan 2026

TechNews reports on the bank's cross-sector anti-fraud network with virtual asset exchanges. It says the bank adopted AI to analyse transaction behaviour, amount and timing. It gives bank-stated figures: over NT$50 million returned to victims by October 2025, over NT$300 million of suspicious transactions blocked by its own controls, and 70% of flagged accounts turned into alert accounts within 90 days.

Full details are in the original sources.

Added 2 Nov 2026 · Updated 2 Nov 2026
Open case page ↗