"We have developed an AI decision model called 'QEV.' While previous approaches involved devising detection methods by defining patterns for phone numbers, address books, and resident registration numbers in code to classify types, this model is characterized by its ability to detect sensitive information and anomalies by analyzing context through self-learning based on a Large Language Model (LLM)."
Kim Sam-young, CTO of QueryPay, stated as much during an interview with Money Today on the 10th, saying, "Various methods have been utilized, ranging from regular expressions and keyword rules to statistical and machine learning classification. However, QEV overcomes the limitations of existing methods that identified sensitive information only within the scope of pre-defined or learned types."
The company explained that QEV is the first domestic product in the category of AI decision models. It is a model specialized for security in the Data Loss Prevention (DLP) field, which prevents the leakage of corporate sensitive information. The company has moved to secure technological leadership by reflecting recent corporate demand regarding data breaches.
According to CTO Kim, QEV analyzes input sentences to classify information and determines its own optimization. Unlike general generative AI, which focuses on answering questions or creating new content, QEV judges whether given information meets specific conditions and presents an appropriate result among predefined options along with the probability for each option.
CTO Kim said, "QEV can determine whether a sentence contains personal information or internal corporate secrets if an employee inputs customer data or undisclosed business materials into an AI service." He added, "It will be possible to identify situations before sensitive information leaks externally and to conduct reviews and responses in accordance with security policies."
QEV implements proprietary decision-making AI technology based on the open-source AI model 'Gemma 4.' Through additional training and optimization using corporate security business data, the company developed a product specialized for sensitive information detection.
CTO Kim stated, "Because QEV can self-learn from each company's data and business characteristics, it is expected to be used in various environments." He added, "The significance of this achievement lies in independently developing a model with State-of-the-Art (SOTA) level performance in the DLP field by utilizing QueryPay's proprietary technology and security-specialized data for additional training and optimization based on an open-source AI model."
As a result of QueryPay verifying performance using internal documents in Korean, English, and Japanese, QEV's sensitive information detection rate was 99.69%. It identified 1,277 out of 1,281 documents containing sensitive information. The detection rate of JEV API, an external commercial decision-making model, was 96.41%, which was 3.28 percentage points lower than QEV.
CTO Kim revealed that QEV has also secured operational efficiency. Operating AI models requires significant computational resources and GPU memory; however, the QEV family of models reduced GPU memory usage by 27.5% to 45.2% while generally maintaining sensitive information detection performance.
QueryPay plans to release QEV on 'Hugging Face,' an AI model sharing platform. It has also prepared an environment where users can easily experience QEV's decision-making capabilities through a separate demo page. The demo service supports Korean, English, Japanese, and other languages. Users can check the model's classification results and probabilities for each option using pre-provided test cases or sentences they input directly. By opening the service to allow anyone to verify the model's performance and potential applications, the company has demonstrated confidence in its product.
CTO Kim said, "Based on this DLP performance verification, we will continue to advance AI technology that companies can utilize in actual business operations and expand the scope of application not only to security but also to various enterprise areas." He added, "The range of applications will be endless."
QueryPay is an enterprise security and AI specialist company established in Silicon Valley, USA, aiming for a domestic Initial Public Offering (IPO). In June 2020, engineers from Kakao and Naver, along with experts in various fields, officially launched the 'Access Control Platform (ACP).' Since then, it has grown in the areas of access control and auditing across data, infrastructure, and business systems. Recently, by adding AI features to its security foundation, the company supports companies in consistently maintaining permissions, approvals, audits, and sensitive information protection while utilizing AI tools and AI agents.
in STOCK
