COONTEC MEDIA COVERAGE

COONTEC and ETRI Begin Development of ‘Obfuscation Technology’ for On-Device AI Security

COONTEC announced on the 16th that it is carrying out a project to develop AI obfuscation technology aimed at strengthening security in On-Device AI environments, where AI computations are performed directly on devices without relying on the cloud. The project is supported by Korea’s Ministry of Science and ICT and the Institute of Information & Communications Technology Planning & Evaluation (IITP).

The project is being conducted in cooperation with the Electronics and Telecommunications Research Institute (ETRI), with the goal of protecting On-Device AI models, which are increasingly used across various industries, from unauthorized leakage, duplication, and reverse engineering.

As the adoption of On-Device AI expands, the importance of AI model security is rapidly increasing. In the past, computing performance and functionality were key competitive factors, but today, maintaining the confidentiality of models and executable code has become essential to ensuring reliability. In particular, in AI applications that handle personal or sensitive data, theft or unauthorized reuse of AI models can lead to serious business and legal risks.

Existing security approaches have shown clear limitations. For example, watermarks used to prove ownership can be altered, while simple encryption methods are vulnerable to memory exposure at runtime and cannot guarantee complete protection. Hardware-based protection and software encryption techniques also face challenges in widespread adoption due to performance degradation and deployment constraints.

To address these issues, COONTEC is developing obfuscation technology designed to prevent external leakage during the actual execution of AI models and code. The technology applies not only to AI models but also to the entire execution structure, including pre-processing and post-processing stages, making it more difficult to analyze and duplicate files and processing logic. It also provides options for balancing obfuscation strength and performance so that it can operate smoothly in real-world commercial product environments.

Through this research, COONTEC plans to fundamentally reduce the risk of exposing AI model structures, weights, and execution logic through reverse engineering, static analysis, or decompilation, while providing an optimal balance between security and performance tailored to each customer environment. The company expects the technology to offer a practical solution for On-Device AI security and enable more trustworthy AI services across industries.

Hyukjun Bang, CEO of COONTEC, said, “On-Device AI is expanding across a wide range of industries, but security vulnerabilities still remain. Through the development of this obfuscation technology, we aim to strengthen both AI asset protection and product reliability, while establishing a new standard for On-Device AI security.”