Giving AI a Headache: Acoustic Adversarial Attacks to Computer Vision Applications

This research presents a novel approach to acoustic adversarial attacks on computer vision applications by utilizing lower frequency sound waves (<20 kHz) to induce physical motion in cameras, leading to misclassification and hallucination in AI models like YOLO11. The study details physical experiments demonstrating the effectiveness of these attacks and identifies key factors that increase vulnerability in AI systems. This work is significant for practitioners as it highlights potential security risks and informs the development of mitigation strategies for robust AI deployment in real-world applications.