Voice AI Systems Are Vulnerable to Hidden Audio Attacks

# Voice AI Systems Can Be Secretly Hijacked With Sounds You Can't Hear Researchers have discovered that AI voice assistants and smart speakers can be tricked into following hidden commands embedded in audio that humans can't detect, like a silent dog whistle for machines. These invisible attacks work 79-96% of the time and can force devices to search the web, download files, or send emails without the user knowing it's happening. The vulnerability affects major AI voice systems from companies like Microsoft and Mistral, raising serious security concerns as these tools become more common in homes and workplaces.