The Google Home Mini is one of the more exciting, and better priced, hardware announcements from last weeks Made by Google 2017 event, but a bug with pre-prod units had them listening to everything you had to say, at least until Google fixed the issue.
The bug was discovered by Artem Russakovski from Android Police, who noted that his Google Home Mini review unit was responding to his TV without activating his Home Mini. A check of his Google My Activity portal showed many requests across days since he activated the review unit.
After reporting the issue, it appears Google is right onto it, diagnosing a faulty touch panel on the device – the Google Home Mini can be activated by either touch or hot (Ok or Hey Google) words.
The wrap-up is that Google reacted quickly, driving up to Artems house to swap out the faulty unit, and then issuing a software fix to disable the touch interface for all remaining Google Home Mini units in the wild (including those given away at the popup donut shops. They’ve also, according to their support page, removed data from the My Activity portal from between the 4th to 7th of October for the Home Mini.
As a leader in search and keeper of many pieces of personal information Google is ultra-conscious of any perceived privacy incursions from their products. Google Glass was notoriously badly received mainly for the on-board camera, and the Google Home (and Mini) come with mute switches for the mic built-in.
Google’s response time for the issue was quite fast, working over a weekend to resolve the issue, so hopefully this will be the end of this.
True and not true. Actually, all Google Homes (as well as Echos and phones with hotwords) listen to everything you say! Thats how they pick up the trigger word. What Adam found is that some Home minis were triggering for no explainable (at the time) reason! When they triggered (wrongly) they did EXACTLY what they were meant to do when triggered! They sent the recording to “Homebase” for handling. As you say, Google fixed the incorrect triggering issue and now Homes of all sizes go back to listening to everything you say! I really think the take away of this… Read more »
The key difference though is that this faulty unit was parsing audio and uploading requests at all times, where as they’re meant to only do this after recognising the trigger phrase. The always-on trigger phase “listening” is done offline to minimise privacy concerns.
My point exactly. It only sent things to “homebase” when triggered. The issue was false/phantom triggering not the functionality of it sending things.
I know it is a Symantec thing but this is an important one to make!