Acoustic adversarial attacks can degrade computer-vision systems
arXiv·medium signal
This paper (arXiv 2606.14658, cs.CV/cs.AI) demonstrates cross-modal adversarial attacks where crafted acoustic signals interfere with AI-driven computer-vision applications. It widens the attack surface practitioners need to model: a vision pipeline can be degraded through an audio channel it never explicitly processes, which matters for anyone deploying CV on shared hardware or sensor-rich edge devices.