Skip to content

Gradient-based defensive poisoning against system identification attacks on cyber-physical systems

Gradient-Based Defensive Poisoning Against System Identification Attacks on Cyber-Physical Systems

arXiv: 其他 AI 安全论文· arXiv:2611.24662· Gonçalo Fernandes, José Cecílio, Victor Mattos, Daniel S. Menasché, Amit Bhaya, Alan Oliveira de Sá·Indexed by this site · 13 hr ago· Originally publishedDefenses· Its place in the scan record →
AI summary

Researchers propose gradient-based defensive poisoning, injecting small perturbations into the measurement signals observed by an attacker to reduce the modeling accuracy of ANN-based system identification attacks against cyber-physical systems. The perturbation aggregates the input gradients of multiple independently trained identification models; on a simulated networked DC motor control system, at the same root-mean-square magnitude it induces larger prediction errors than random poisoning. The defense targets decoy environments such as CPS honeynets and also degrades N4SID-based system identification, indicating transferability.

Read originalarxiv.org