Financial Security Institute and Three Internet-Only Banks Join Forces... Launch of Joint AI Defense Network Against Voice Phishing and Fraudulent Accounts

Federated Learning-Based Joint AI Model to Launch in July
Detection Performance Improves by Up to 205%
Expansion to Small and Medium-Sized Financial Institutions Planned for Fourth Quarter

The Financial Security Institute announced on June 24 that, in collaboration with KakaoBank, Toss Bank, and K-Bank—the three major internet-only banks—it has jointly developed a “voice phishing detection artificial intelligence (AI) model,” which will be deployed in real-world operations starting in July.

This model was developed by applying a federated learning technique, which allows each financial institution to share and integrate AI models trained on their own data without exposing any original training data externally. The key feature is that it creates an AI that can be used jointly while safeguarding each bank’s personal information and security.

This project combined the on-the-ground experience of the three internet-only banks, which have directly developed and operated voice phishing detection AI, with the Financial Security Institute’s research and federated learning technology. The fraud detection capabilities accumulated by each bank are now able to be shared across institutions, enabling the detection of fraudulent transactions that individual banks previously could not identify. Performance verification of the joint model showed that its detection accuracy improved by up to 205% compared to individual models.

The technological advancements accumulated through the development of the joint voice phishing detection AI model have been recognized internationally, with the model being accepted by CIKM and NeurIPS, two of the world’s most prestigious AI conferences, thereby demonstrating the competitiveness of Korea’s financial AI technology.

Starting in July, the joint model will be applied in the field alongside the three internet-only banks’ existing AI models and their Financial Fraud Detection Systems (FDS).

The Financial Security Institute is also currently working on developing joint models for other financial sectors, such as commercial banks and card companies, where transaction data structures are similar. The institute plans to gradually expand the use of these joint models across the entire financial industry.

Additionally, in the fourth quarter of this year, the joint model will be incorporated into the Financial Security Institute’s ASAP (AI platform for sharing and analyzing information on telecommunications and financial fraud), enabling its use by smaller and mid-sized financial firms in the secondary financial sector. As a result, financial institutions with relatively limited AI development capabilities and data resources will also be able to easily utilize advanced voice phishing and fake account detection technology.

Park Sangwon, President of the Financial Security Institute, stated, “Building a collaborative system across the entire financial sector is crucial to respond to increasingly sophisticated and organized voice phishing crimes. We will not only continue to develop joint voice phishing detection models, but also leverage the ASAP platform for proactive and preventive detection, thereby strengthening consumer protection in the financial sector.”

This content was produced with the assistance of AI translation services.