When people think about cyberattacks, they usually imagine malware or software vulnerabilities. However, some attacks target the physical behaviour of electronic devices instead. By briefly disturbing essential parts of the device under attack, attackers can sometimes force it to make mistakes and thus bypass security measures. These attacks are becoming an important concern in hardware security.
In our project, we investigate how Artificial Intelligence (AI) can help detect such attacks by analysing the device’s behaviour.
How we captured attack data
For our experiments, a commonly used hardware security research platform was chosen. This platform allows researchers to precisely perform an attack while measuring the targeted device’s behaviour.
By repeating these experiments many times, we created a dataset of different scenarios. This dataset serves as the foundation for training an AI model to automatically identify anomalies in the device’s operation.
Teaching AI to recognize abnormal behaviour
Our approach focuses on training a deep neural network on distinguishing between modes of operation. Rather than analysing software activity, the AI evaluates measurements and learns to identify subtle patterns that appear when a device is being physically manipulated.
Potential applications
AI-based anomaly detection could improve the resilience of many types of systems, including:
- Secure embedded devices
- IoT products and smart home technology
- Automotive electronics
- Industrial control systems
- Medical devices
By detecting abnormal behaviour early, devices could react to potential attacks before security is compromised.
Towards more resilient hardware
This work shows how AI can be used not only for data analysis or automation, but also for improving the security of electronic systems. By enabling the identification of abnormal behaviour caused by physical attacks, such approaches could contribute to the development of more resilient and trustworthy electronic systems in the future.
Blog signed by: UZL team