What is it about?
We designed a Generative Model, with a customized loss function and training algorithm, to automatically generate cyber-attack scenarios for evaluating the vulnerability of complex control and infrastructure systems. Instead of relying on detailed physical models or pre-collected attack data, this data-driven generative model learns how to produce stealthy and effective attacks through interaction with physical systems.
Featured Image
Photo by Debbie Ducic on Unsplash
Why is it important?
We provide a scalable, data-driven method to explore the full vulnerability space of cyber-physical systems—far beyond what traditional model-based or single-attack methods can capture. It gives defenders a way to automatically generate realistic, stealthy cyber-attacks, helping them understand and strengthen system vulnerabilities before adversaries exploit them. Additionally, it allows critical infrastructure operators to proactively test resilience using AI-generated attack scenarios that mimic real-world adversarial behavior.
Perspectives
Working on this paper was especially rewarding because it brought together ideas from control theory, machine learning, and cybersecurity in a seamless way. Seeing how a generative model can uncover vulnerabilities that even domain experts might overlook reinforced for me just how powerful data-driven methods can be in protecting critical infrastructure.
Yu Zheng
Florida State University
Read the Original
This page is a summary of: Generative Vulnerability Assessment for Cyber-Physical Systems, ACM Transactions on Cyber-Physical Systems, November 2025, ACM (Association for Computing Machinery),
DOI: 10.1145/3776543.
You can read the full text:
Contributors
The following have contributed to this page
