In case you weren’t already convinced, the Belgian publication VRT NWS has substantial proof that whatever you are saying into the Google Home devices and smartphones with Google Assistant can be directly heard by Google employees.
What’s even more alarming is that sometimes Google Home smart speakers and smartphones start recording people even if they haven’t said the activation command “Okay Google.” This might be a glitch or at times users may unintentionally wake up the digital assistants on their device. Either way, this allows people who for work for Google to listen in on private conversations, some of which have sensitive information too.
Not too long ago, it was reported that Amazon outsources transcribing jobs, allowing contractors to listen to audio recorded by Echo speakers. Thus VRT NWS decided to find out if Google is doing the same.
Google has clearly stated in its terms and conditions that everything that is said to Google Assitant and Home devices is recorded. However, nowhere does the company say that human employees have access to those recordings.
A person who works for a Google subcontractor has revealed that thousands of employees all over the globe can access a secret feature of the company’s free online too Crowdsource, which contains audio excerpts. The job of the workers is to analyze those audio clips.
Google has admitted that it takes help from language experts to improve its voice recognition technology. However, the search giant says that only about 0.2 percent of audio files are analyzed and that this work is crucial for sustaining Google Assistant.
To be fair to Google, it will probably not be able to process voice commands without improving its AI system and currently, this cannot be done without some degree of human involvement. However, the company should explicitly mention this on its website.
Perhaps the bigger issue is that although the company says that none of those audio clips are linked to any identifiable information, VRT NWS, which went over a thousand such recordings, said that some of the clips also reveal sensitive information such as addresses and using that data, they were able to track down the people involved.
In some cases, to gain a better understanding of what’s being said, the transcribers also look up names and workplaces mentioned in the voice clips on Facebook and Google and this lets them identify the people involved easily.
What’s even more worrisome is that about 153 of those 1000 recordings were not meant to be recorded in the first place. This means that Google was able to access and store sensitive audio clips even when the words “Okay Google” were not said, unbeknownst to the people being recorded.
Some of the recordings also contain health data and this may land Google in trouble, as EU’s GDPR mandates explicit consent for collecting such information. Ideally, Google should be more open about how its system works, but it is unlikely to do that, as some of the steps involved might creep out users. For now, a spokesperson has said that the Mountain View-based company will review how it can provide clarity to its users regarding how the data is used to improve its speech technology.