Hackers can talk to your devices with inaudible voices and there is nothing you can do.

They can call numbers, open websites and even order items on Amazon.

A recent paper from students are the University of Zhejiang (China) shades some lights on the security flaws of Artificial Intelligence assistants on your phone. We know that AI and humans don’t work the same way when it comes to Speech Recognition, however nobody ever successfully exploited these differences.

Apparently, one of those differences is that AI can pick up on sound levels inaudible to humans.

It’s called a Dolphin Attack, and it looks scary:

The scariest part is how this attack can propagate, it is easily embeddable in a music file, a youtube video or over speakers in a crowded area.

Google, Amazon and Apple have already taken steps to mitigate this attack by reducing the sounds frequency range the AIs listen for. Authors of the paper has also suggested that is it possible to detect those attacks using Support Vector Machines (SVMs). But my guess is that we will see a lot more attack vectors against AIs in the coming months.

Source DolphinAttack: Inaudible Voice Commands, 31rst August 2017, https://arxiv.org/pdf/1708.09537.pdf