Researchers have found it’s possible to send commands to voice-activated assistants like Apple’s Siri and Amazon’s Alexa that are undetectable to the human ear, essentially hacking them and opening users to risk, The New York Times reports.
A group of students from the University of California Berkeley and Georgetown University published a paper this month showing they could embed commands directly into music recordings or spoken text. For example, while a human may hear music or spoken text, Amazon’s Echo might hear a command to add something to their shopping list.
The speech recognition systems used by phones and voice assistants generally translate sounds to letters and build those into words and phrases. By making small changes to audio files, researchers were able to cancel out the sound that the speech recognition system was supposed to hear and replace it with a sound that would be transcribed differently by machines.
Nicholas Carlini, a fifth-year Ph.D. student in computer security at U.C. Berkeley, says that so far the secret commands have only been proven in their labs, but it could be exploited by others.
More than half of all households will have at least one smart speaker by 2021, according to Juniper Research, and most smartphones already have some type of vocal command system. Amazon and Google both say they are taking security measures to protect users, though they don’t disclose specifics. Apple says its smart speaker, HomePod, is designed to prevent commands from doing things like unlocking doors and that Siri can’t be accessed unless a phone or tablet is unlocked.
