<p>This paper proposes an amplitude-controllable multiscroll system, which consists of a single nonlinear term and six linear terms. The system’s mathematical equations are concise, and its circuit implementation is straightforward. Based on this system a chaotic circuit with external environment sensing capability is designed to respond differently to illuminance, temperature and sound intensity in the environment. Subsequently, in order to further enhance the complexity of the system, Julia fractal was introduced into the system and an “attractor shape flip” phenomenon was found during numerical simulation. Finally, a chaos adversarial attack algorithm is designed based on the Projected Gradient Descent (PGD) algorithm, driven by an amplitude-controllable multiscroll chaotic system and Julia fractal, named Chaotic-PGD. Through a comparative analysis of perturbation characteristics, attack performance, and cross-classifier attacks, the study shows that, compared to the traditional Classic-PGD algorithm, the adversarial samples generated by the Chaotic-PGD algorithm exhibit increased diversity, higher concealment, and more cross-category errors. As a result, Chaotic-PGD excels in diversity and unpredictability, enhancing its stealth and effectiveness in adversarial attacks. This characteristic provides Chaotic-PGD with a unique advantage in simulating complex attack scenarios.</p>