Contractors hired by Google have been listening to audio recorded by its virtual assistant—even files recorded after the “OK, Google” wake word wasn’t invoked—reports Belgian outlet Vrt Nws.
According to the story, the news outlet listened to many recordings, some that didn’t start with Google’s wake word, containing personal conversations, professional phone calls with private information, interactions between kids and their parents, and other sensitive information.
In an email to Fortune, Google noted its workers are listening to improve the system. According to the search giant, users can choose to have their audio recordings deleted, and the virtual assistant can be used without opting in to “Voice and Audio Activity” (the setting that allows Google to store user audio). The company also addressed its devices recording audio without an “OK, Google” prompt in a blog post, saying there are “a number of protections in place” to prevent a false positive from triggering.
Google isn’t alone in letting employees listen to users audio recordings. In April, Bloomberg reported that Amazon let its employees listen in on users of its Alexa always-listening, virtual assistant. According to Amazon, the device only records after it hears the wake word, “Alexa.” Amazon workers listened to “as many as 1,000 audio clips per shift” in an effort to improve the virtual assistant, the report noted.
As when Amazon seemingly overstepped its data-gathering bounds, some privacy experts are concerned about Google’s recordings. “This data collection, and the related incident with Amazon employees listening to Alexa recordings, aren’t mistakes at all,” said Christo Wilson, a computer science professor at Northeastern University. “Both companies will keep collecting and analyzing audio data in this way, regardless of the news fallout, because they have to. It’s the only way to keep improving their voice assistant technologies.”
Google’s history of overzealous data collection
Many Google Assistant users are aware that everything said after “Ok, Google” is uploaded and examined. Putting other, everyday audio under the same microscope, however, may not sit well with some users. But Google has a history of collecting data that users and privacy advocates found to be overzealous.
For example, Google Map’s street view cars were found in 2010 to have been accessing nearby open Wi-Fi networks and vacuuming up unencrypted data over years across multiple countries. In response, Google said it collected this data by mistake. It also said listening in on unencrypted Wi-Fi networks wasn’t illegal, likening the act to listening to public AM/FM radio stations. In 2013, the company lost in federal court and was held responsible for damages.
Google was also caught tracking the location of Android users in 2017, even if users specifically turned off location services on their device. Smartphones with Google’s operating system installed would collect the address of nearby cell towers and relay that info back to Google.
More recently, Google’s Chrome web browser was cited for forcing users to log in with their Google account in 2018. In version 69 of Chrome, users that logged into a Google website would also see that account attached to their web browser—meaning a users’ browsing habits were attached to their instance of Chrome. Previous versions of Chrome had a specific login section for users who wanted to do so, but this new version automatically attached users’ profiles to their browsing activities. Google reverted to its previous account association following user outcry.
While Google asserts that its assistant’s data collection is being used to bolster virtual assistants systems, a huge portion of the company’s revenue comes from collecting lucrative information for its online advertising business.
And the problem, to privacy advocates like Wilson, lies in users not knowing how much of their data is being tracked, nor where it’s going. “Any device or service that collects data for advertising purposes ends up sharing that data with hundreds of third-parties in the advertising ecosystem,” he says. “The best disclosure you will find for this behavior (if you find any at all) is that ‘data will be shared with third parties.’ What data? With whom? For what purposes?”
Over the years, Google’s posture has bent toward “collect data first, ask questions later.” With Google’s assistant products, this stance is manifesting itself in the company examining audio that users reasonably expect to be collected as well as recordings that Google said it wasn’t collecting to begin with.
While the assistants are virtual, there may be more humans involved in the process than we realize who have sweeping access, Wilson warns. “The companies have done a very good job of making people believe that these new, smart technologies are entirely mediated by machines,” he says. “But the dirty secret is that there is a small army of people behind the scenes analyzing, curating, and moderating the content.”
More must-read stories from Fortune:
—What people get wrong about artificial intelligence and China
—A new A.I. is running the table against poker pros. Is business strategy next?
—How IBM is fine-tuning its Wimbledon tech
—For Uber riders, guaranteed comfort, comes at a cost
—Listen to our new audio briefing, Fortune 500 Daily
Follow Fortune on Flipboard to stay up-to-date on the latest news and analysis.
