Peng Jinlong: Taiwan’s banks’ AI fraud prevention efforts are effective! The number of alert accounts has "negative growth for the first time", but what about the public complaints about accidentally locked accounts?

Taiwan’s AI fraud crackdown has been somewhat effective, with the National Bank of China warning that the number of accounts showed negative growth for the first time in September. However, a large number of cases where salary transfers, tuition fees and mortgage transfers were mistakenly locked have also emerged simultaneously.
(Preliminary summary: Taiwan’s special anti-fraud law will include “crypto exchanges” in the joint defense! The Financial Supervisory Commission announced that the draft of four sub-laws will be released before the end of the year)
(Background supplement: Anti-fraud! Securities accounts in domestic banks have been frozen “please go to the branch to unlock”, netizens: Who will pay for defaulted delivery?)

Contents of this article

In Taiwan, where fraud techniques are endless, AI It is being regarded as the anti-fraud shield of the new generation, but it has also accidentally created another scar. On October 13, Peng Jinlong, chairman of the Financial Supervisory Commission, stated in the Legislative Yuan that in September, the number of China National Bank warning accounts showed negative growth for the first time, showing that the bank has achieved good results after introducing AI to combat fraud. However, a large number of cases where salary transfers, tuition fees and mortgage transfers were mistakenly locked have also emerged simultaneously.

Alert accounts have negative growth for the first time

According to statistics from the Financial Supervisory Commission, the number of alert accounts has soared from 66,000 at the beginning of 2022 to 150,000 at the beginning of 2024. However, after the growth dropped below 1,000 households in June this year, the growth turned negative for the first time in September. This data shows that the AI model is indeed effective in filtering high-risk accounts and has also alleviated the long-term fraud prevention pressure faced by banks.

Wrong locking of people is the "friction behind the scenes"

However, good results also come with a high price. Legislator Wang Yumin pointed out during questioning that the bank's AI detection was too loose, causing normal customers to be included in the "high-risk" list. Take China CITIC Bank as an example. In the past, after detecting a short-term surge in the number of salary transfers, it froze the accounts of many third-party payment users, forcing people to go to the counter or call customer service for several hours to unlock them. For some families, the inability to have their salaries credited to their accounts on time and their mortgage payments to be withheld will seriously affect their normal lives, thus increasing public complaints.

Policy fine-tuning: Four major requirements of the Financial Supervisory Commission

However, under the pressure of public opinion, Peng Jinlong still emphasized that his stance on combating fraud will not shrink. He publicly stated: "Control will not stop, that is what the fraud group hopes most." However, in order to reduce accidental injuries, he also proposed four major measures for banks. First, the feedback model is based on actual mislock cases, allowing AI to make more accurate judgments on transactions. Second, it is mandatory to set up a 24-hour dedicated line and authorize frontline personnel to complete simple unlocking within one hour. Third, introduce the risk classification (RBA) model to avoid the "one size fits all" phenomenon. Fourth, share cases and best practices through the Banking Association platform to form an inter-bank joint prevention mechanism.

In addition, Peng Jinlong also distinguished between two freezing situations, emphasizing that if the order is for inspection and investigation, the bank has no right to unfreeze; if it is the bank's preventive management, it needs to be processed quickly.

Future Challenges: Balancing Technology and Human Rights

AI detection will rely on a large amount of transaction data, which will inevitably involve privacy issues. Although warning about negative account growth is a big progress, how to continuously optimize the model and shorten the unlocking process is the focus of continuous optimization by financial institutions. Only by allowing technological tools to truly return to their original intention of taking care of users can anti-fraud and civil rights move forward side by side on the same axis.