Silent Sabotage: Inaudible Audio Attacks Can Hijack Voice AI Behavior

Research published in IEEE Spectrum reveals a disturbing vulnerability in Voice AI systems: they can be controlled by sounds that are completely silent to the human ear. These “hidden audio attacks” exploit the way AI models process acoustic data, allowing malicious actors to embed commands within music or ambient noise that bypass human detection.

Studies demonstrate that these inaudible signals can trigger serious actions, such as unlocking smart doors, making unauthorized purchases, or accessing sensitive websites. As voice-controlled interfaces become ubiquitous in homes and offices, this research underscores the urgent need for a new generation of security protocols that can distinguish between legitimate human speech and sophisticated adversarial acoustic attacks. The race to patch these silent vulnerabilities is now on.