Detecting RISC-V hardware attacks using instruction opcodes with QEMU

Abstract

Póster presentado en HiPEAC 2026 (Cracovia)

As malware becomes more advanced, there is a growing need for more accurate and robust detection systems. Artificial Intelligence (AI) techniques, particularly those based on machine learning and deep learning, have emerged as powerful tools for identifying and classifying malicious behavior. Can we detect RISC-V Hardware Attacks using a machine learning model analyzing sequences of executed instructions? This work builds a suitable dataset, trains machine learning models and evaluates models in 2 scenarios: known and zero-day attacks, showing that known attacks can be detected with a high accuracy, while zero-day attacks may be detected but depends on the similarity with other trained attacks.

Funded by the European Union. Project number: 101093062.

Preprint