Background information

A Swiss Alexa: are we witnessing the dawn of a massive eavesdropping campaign?

Dominik Bärlocher
29.4.2021
Translation: Patrik Stainbrook

Amazon Alexa is upon us. The voice assistant has officially been launched in Switzerland. However: is she always listening? Or, in other words: how much does Alexa monitor you?

Amazon has brought Alexa to Switzerland. The smart voice assistant lives in your speakers and smartphones, all the while listening quietly whenever you say something. A massive concern in terms of data protection: Jeff Bezos is always there. He’ll know what you say to your cat and what’s being discussed at dinner. He’ll even be creeping on you during sex. Horrifying!

It’s hysteria and rational apprehension all in one, every smart device will spark new discussions.

Which I’m certainly very happy about.

Data protection and privacy are being threatened by a new foe practically every week. This app wants more data, that one has you revealing something about yourself again. You’re trapped by the classic Internet warning, especially when it comes to free services such as Amazon’s Alexa: If it’s free, then you’re not the customer, you’re the product.

Is Alexa always listening in?

The Alexa app provides an initial indication of what exactly Alexa is listening to. You’re free to inspect every one of your voice commands Alexa has interpreted. The reason? If Alexa gives you an incorrect answer, you can say «No Alexa, you messed up» via the app, helping Amazon improve their service. Perpetually optimising the product harvested from you. This also allows Alexa to send your voice recordings overseas for analysis.

On a technological level, Alexa only listens when she hears the name «Alexa» after a short period of silence. So when you mention her name in conversation, she doesn’t respond.

After all, every device with a voice assistant has two separate «hearing circuits» installed, i.e. two systems that listen to you. They are named as follows.

  1. Passive
  2. Active

The passive system is only there to respond to «Silence, Alexa.» No data is transferred. The passive system is local and read-only on Sonos One, as far as I can tell. According to several sources, you can change the trigger word on Echo devices. So if you have a cat named Alexa, then you need to rename either the cat or Amazon’s Assistant. You can’t do that on Sonos One.

When the passive system reacts, it triggers the active system. This is the part that records your voice and analyses it in the cloud. You can tell visually from the devices themselves. On Echo devices, a blue ring lights up, indicating «Attention: recording in progress.» However, since Alexa responds to all voices that say «Silence, Alexa», this can be fairly easily overridden.

My Sonos One Alexa activated at around 47 seconds into the above video. Then again at 1:04. Turns out Alexa has extremely sharp hearing. It’s stood in my living room, a few metres and doorways away from my bed. I can nevertheless give commands at normal conversational volume, and Alexa responds correctly. How well Alexa hears depends on how good her microphones are. It’s quite possible that Sonos has installed better microphones than Amazon in its Echo Dots.

So if I record a voice command of me saying «Alexa, buy a Breguet» and run it in your home, Alexa will respond, assuming you have all the purchase and delivery information correctly on file with Amazon.

Alexa doesn’t always listen.

But she listens to her name. So if you have a cat named Alexa, you’ll already have a problem. Or if you yourself are called Alexandra, nicknamed Alexa.

The amount of data collected

There are currently 100 million, or 100,000,000, Alexa devices worldwide according to Marketingland. Now, if these devices were listening 24 hours a day, they would quickly amass an enormous amount of data.

  • One minute of MP3 audio in 128 kb is exactly 1 MB.
  • A day contains 24 hours, times 60 minutes gives us 1440 minutes.
  • One user would produce 1440 MB of data – daily.

Accordingly, all Alexa devices would produce 144,000,000 MB of data. Every. Day. That would be 144,000 terabytes or 144 petabytes. Let’s compare: statistically, your laptop probably has about half a terabyte of memory.

Amazon would then have to listen through this incredible amount of data, or so the myth goes. According to some horror stories, this is actually happening right now. Conjuring up dystopian images of huge offices filled to the brim with employees, all wearing headsets, just listening to what users say and taking notes of what they hear. Just in case anything noteworthy pops up.

They could manage more if Amazon filtered out silence via artificial intelligence.

Another person, listening in… maybe

If an Amazon employee is listening, here’s how it goes, according to Time Magazine’s sources: in nine-hour shifts, employees around the world transcribe recordings and feed them back to Alexa. To do this, the software receives annotations from the transcribers so it can better understand what exactly happened.

This army of listeners isn’t only directly employed by Amazon, but can be provided by contractors. These employees and contractors work in the USA, Costa Rica, India and Romania. Per shift, a person hears about 1000 recordings, all following the «Alexa» command.

One employee reports spending hours on the words «Taylor» and «Swift.» They taught Alexa’s software when a user produced «Taylor» – either the name or a homonym for «tailor» – or «Swift» – as in «fast» or «hurried» – in a different context, or whether the user was actually referring to singer Taylor Swift.

Other employees report an internal chat used to share funny or horrible recordings. Sometimes it’s «a woman singing badly off key in the shower», other times it contains «a child screaming for help». Amazon has specific guidelines on how employees should handle hearing something like a cry for help or conspiracy to commit a crime. Two employees from Romania say, «[…] it wasn’t Amazon’s job to interfere».

Couldn’t Amazon do a lot of good with an emergency monitoring device? Definitely. But interacting with your life outside of explicit commands may not be in a large corporation’s wheelhouse.

Amazon itself responded to Time Magazine in 2019. While people are listening, the group has no information about who the people in the recordings are, they said. Data is anonymised.

In the app’s settings, you can turn off data sharing for analytics and development purposes. Which I recommend you do. This is, by the way, better than the «This call may be recorded for training purposes» line you’ll hear out of any call centre. With them, you only get to choose between «Monitored customer service» or «Leave me alone, dude.»

Why you still get ads

Another big but: you’ll often receive advertisements about things that you mentioned in conversations with your mother, let’s say. Or referring to that new cat toy you mentioned to your pet at dinner. Your cat answers with «Meow», then the Internet butts in with «Buy now!» However, this isn’t due to your smart speaker, be it Alexa, HomePod or Google Mini, but a completely different mechanism.

After all, Alexa isn’t the only thing spying on you. The Internet as a whole has been checking on you for years. Third-party cookies and their successors analyse your browsing behaviour. Apple recently started putting a stop to this, and did Google.

By the way, the cover image above is from srlabs.de.

95 people like this article


User Avatar
User Avatar

Journalist. Author. Hacker. A storyteller searching for boundaries, secrets and taboos – putting the world to paper. Not because I can but because I can’t not.


Smart home
Follow topics and stay updated on your areas of interest

Security
Follow topics and stay updated on your areas of interest

Background information

Interesting facts about products, behind-the-scenes looks at manufacturers and deep-dives on interesting people.

Show all

These articles might also interest you

  • Background information

    We love to lüft: the very German and Swiss art of airing a room

    by Katherine Martin

  • Background information

    Everything’s better these days! Seven reasons why family driving holidays are becoming more and more enjoyable

    by Michael Restin

  • Background information

    «Data from wearables means you can skip the sleep clinic»

    by Martin Jungfer