{"id":52398,"date":"2016-08-31T15:57:42","date_gmt":"2016-08-31T22:57:42","guid":{"rendered":"https:\/\/blogs.mcafee.com\/?p=52398"},"modified":"2025-08-26T23:30:01","modified_gmt":"2025-08-27T06:30:01","slug":"voice-recognition-hack","status":"publish","type":"post","link":"https:\/\/www.mcafee.com\/blogs\/internet-security\/voice-recognition-hack\/","title":{"rendered":"When Voice Recognition Goes Wrong: YouTube Videos Can Say \u201cOk Google\u201d"},"content":{"rendered":"

YouTube videos are like today\u2019s candid camera: they can be fun, educational, and, sometimes, a window into someone else\u2019s most embarrassing moments. But videos we watch could just as easily flip the tables and allow intrusions into our own lives. How? By containing secret instructions<\/a> capable of triggering voice-recognition tools on our smartphones. It just takes a few well-placed words in the video\u2019s audio, and our devices can be ready to obey. After all, it only takes \u201cOk Google.\u201d<\/p>\n

Researchers at the University of California, Berkeley and Georgetown University demoed this attack<\/a> on, well, YouTube. The demo shows how a heavily modified voice command can trigger the capabilities of Google Now, the personal assistant on Android devices. Once activated, the phone\u2019s program can perform a number of functions. According to the scholars who demonstrated the technique, the concern is that these types of sounds can be embedded in videos and on platforms like YouTube.<\/p>\n

This works because smartphones process sound differently from a human ear. By manipulating a sample of a real voice saying \u201cOk Google,\u201d perpetrators can create a noise that a mobile\u2019s personal assistant recognizes, even if it seems strange to us.<\/p>\n

Consequently, cybercriminals could, potentially, give orders to a smartphone without the user\u2019s knowledge. One of the researchers, Micha Sherr, demonstrated this technique by opening his Facebook app during an interview with CNBC<\/a>. That\u2019s just the tip of the iceberg. This attack can leverage any command possible with vocal-recognition. For example, researchers believe this method could prompt a smartphone to visit a website containing malware \u2014 potentially without alerting the user.<\/p>\n

Luckily, this attack isn\u2019t likely to become widespread. So far, researchers have only demonstrated its potential \u2013 criminals haven\u2019t actually been discovered using it. Additionally, there are easier methods for crooks to implement on a mass-scale. However, the discovery is poignant, given the prominence of vocal-recognition today.<\/p>\n

As devices become our personal assistants \u2013 whether Android, iPhone, or Amazon Echo \u2013 we will be relying on their capabilities more and more. Consumers should have an awareness of associated safety issues. After all, who wants to worry while watching video compilations of people falling?<\/p>\n

Learn these tips to use voice-recognition safely:<\/h2>\n