Hackers Find That Inaudible Sounds Hidden in Podcasts or Random Videos Can Hijack Your AI Voice Chatbot

Researchers from China and Singapore showed that inaudible adversarial audio can control voice AI models without being noticed by humans.

The hidden audio can be embedded in everyday media like podcasts, songs, or videos, making the attack hard to detect.

The demo targeted open source models used in products from Microsoft and Mistral, exposing weaknesses in current defenses.

The findings raise concerns about voice assistant hijacking, account access, and data theft through ordinary audio content.