1: \begin{abstract}

2: % The data-driven attack strategies recently have received much attention when the system model is unknown to the adversary. Depending on the unknown linear system model, the existing data-driven attack strategies have already designed lots of efficient attack schemes. In this paper, we study a completely model-free attack scheme regardless of whether the unknown system model is linear or nonlinear. The objective of the adversary with limited capability is to compromise state variables such that the output value follows a malicious trajectory. Specifically, we first construct a zeroth-order feedback optimization framework and uninterruptedly use probing signals for real-time measurements. Then, we iteratively update the attack signals along the composite direction of the gradient estimates of the objective function evaluations and the projected gradients. These objective function evaluations can be obtained only by real-time measurements. Furthermore, we characterize the optimality of the proposed model-free attack via the optimality gap, which is affected by the dimensions of the attack signal, the iterations of solutions, and the convergence rate of the system. Finally, we extend the proposed attack scheme to the system with internal inherent noise and analyze the effects of noise on the optimality gap.

3: % Extensive simulations are conducted to show the effectiveness of the proposed attack scheme.

4: % \end{abstract}

5: