Is 'Security Threat' Limited to Hacking? Physical Technology Theft Could Be a Hidden Risk
A new security threat has emerged, raising the possibility that sensitive information can be easily extracted in physical environments without traditional hacking methods. In light of this, experts emphasize the importance of implementing not only cybersecurity measures but also physical safeguards in parallel.
On March 31, KAIST announced that a joint research team led by Professor Jun Han of the School of Computing at KAIST, along with the National University of Singapore and Zhejiang University in China, has developed and demonstrated an attack system named ‘ModelSpy’ capable of stealing artificial intelligence (AI) model architectures from a distance using only a small antenna. Through this demonstration, the team proved the potential for security threats in physical environments.
The core functionality of ModelSpy, developed by the joint research team, is designed to capture subtle signals emitted during AI operation—much like an eavesdropping device—to analyze and reconstruct the internal structure of the AI model.
These subtle signals refer to electromagnetic waves generated by the graphics processing unit (GPU) responsible for AI computations. The research team succeeded in reconstructing the layer composition and detailed configuration values of the model by analyzing the electromagnetic wave patterns emitted by the GPU during AI processing.
In recent experiments on five different GPUs, ModelSpy was able to identify the architecture of AI models with relatively high accuracy—even from behind walls or at distances of up to 6 meters. The system estimated the core structure of deep learning models, namely the layers, with an accuracy of up to 97.6%.
This demonstrates that, unlike conventional hacking which requires direct server intrusion or malware installation, an attack can be launched using only a small antenna portable enough to fit in a bag. This means that significant security threats can arise not just in cyberspace, but also in physical settings.
Recognizing the risk that such technology, if exploited in real-world scenarios, could lead to the leakage of core AI assets from companies, the research team proposed countermeasures such as electromagnetic interference and computational obfuscation. By offering these measures, the team went beyond a simple demonstration of the attack and suggested practical defense strategies.
Professor Han stated, "This research demonstrates that AI systems can be exposed to security threats in physical environments that are different from conventional hacking methods. It highlights the need to establish a 'cyber-physical security' framework that encompasses both hardware and software to protect critical AI infrastructure, such as in autonomous vehicles and national infrastructure."
Hot Picks Today
"Two Spend 500,000 Won in a Flash, Yet Wait Two Hours?" ... 'Dreamland' Shunned by Kids and Foreign Tourists
- The Era of Getting 200 Million Won After 30 Years Is Over... A Japanese Bank's "Perpetual Department Head" Who Just Clocked In Cries Out
- Spain Hit by Its Worst Tragedy... Lee Kangin, Moreno and Real Madrid Join Soccer World in Mourning
- They Said, “Make It Your Home for Life” - Then Suddenly Kicked 200 Passengers Off... What Happened to the Around-the-World Cruise?
- "130 Rice Bowls Sold Out in a Day"... 'Unfiltered Face' Adds 8 Million Won to Night Market Sales: China's Bungeunwoo
The study was led by Professor Jun Han of KAIST’s School of Computing, who participated as a corresponding author. The research findings were presented at the international conference ‘NDSS (Network and Distributed System Security Symposium) 2026’ in the field of computer security, where the study was recognized for its innovation and selected as the best paper.
© The Asia Business Daily. All rights reserved. Unauthorized AI training and use prohibited.