voice ai security

Voice AI security is a growing concern as demonstrated by the AudioHijack attack, a hidden-audio prompt injection technique that can manipulate voice AI systems into following unauthorized instructions. Researchers from Zhejiang University, the National University of Singapore, and Nanyang Technological University presented this attack at the IEEE Symposium on Security and Privacy in 2026. The method exploits the audio channel as a command surface, moving beyond simple speech recognition to tool use. This highlights the need for robust defenses against prompt injection in voice AI, as users cannot reliably inspect audio for hidden commands. The discovery underscores that voice AI security must address new attack vectors where sound becomes a vector for unauthorized actions.