Alexa, I Object.
Tuesday, January 10, 2017
Posted by: Mary Mack
What is uttered in Vegas, stays in Vegas. Or does it?
Alexa, the voice assistant from Amazon, was a star of the Consumer Electronics Show (CES) in Vegas. Boasting over 7000 skills, or voice activated tasks, Alexa is now built into a wide variety of platforms. Voice programmers are heralding the advent of speech-enabled everything.
At the same time, Amazon was served and partially complied with a law enforcement request for Alexa data that police believe contain commands or music streaming contemporaneously with a hot tub murder. The snippets might also tie other evidence together if background noise or conversations surface.
Jim Calloway, Director of the Oklahoma Bar Association's Management Assistance Program, wonders when Alexa data will be requested in civil cases, for example, during a divorce proceeding when one of the parties is an avid Alexa user, with the requester hoping to discover evidence in background or overheard conversation.
Looking at Alexa from an e-discovery perspective, one must first analyze how Alexa works and how Alexa’s data is stored. Alexa, like other voice interfaces, waits for a key word to wake it up to receive instructions. This is sometimes referred to as a “wake word” or activation sequence. For Alexa, the word is Alexa, or it can be changed to Amazon. According to documentation, while Alexa is listening for the wake word, it is not recording. Once awakened or activated, voices and background noise are recorded, encrypted and sent to Amazon servers for interpretation.
According to Craig Ball, who chronicled the dangers of letting Alexa be alone in the kitchen with friends:
“Using the Alexa app on my phone or computer, I can view a list of every interaction since Alexa first came into my life, and listen to each recording of the instruction, including background sounds. You may call this ‘creepy.’ I call it ‘evidence.’”
The files are stored in-part on the local device and in-part on the server at Amazon. More specifically, according to Sean Gallagher of Ars Technica:
“Echo and other devices that use Alexa monitor audio recorded by a microphone for utterances; when speech (or some audio) is detected, it is recorded to a binary audio file and sent in a JSON message back to the Alexa service.”
The audio queries are saved at Amazon until deleted by the user. The queries are used to improve the Alexa speech recognition in product development and testing, as they are at Google. However, trying to extract the data is an exercise of screen capture, play and record, according to Craig Ball, who sees the need for a user controlled “take out” option like Google or Twitter has.
Alexa users can find a running list of their queries in the Alexa app in Settings > History. If a user has several Alexa devices in their arsenal, each one has its own listenable queue of requests.
From a privacy perspective, Apple has designed its speech interface with some privacy by design features, for example, not storing ID’s and emails with the recordings. Alexa, on the other hand, keeps the data unless deleted by the user.
“While Apple logs and stores Siri queries, they’re tied to a random string of numbers for each user instead of an Apple ID or email address. Apple deletes the association between those queries and those numerical codes after six months. Your Amazon and Google histories, on the other hand, stay there until you decide to delete them.”
For information governance and retention, there is a way to bulk delete Alexa data, something not to do when under a legal hold. It is also a user by user exercise.
“The app only lets you do that one entry at a time. To do a bulk-delete, you’ll need to visit amazon.com/myx, click the “Your Devices” tab, select your Alexa device, and click “Manage voice recordings.” A pop-up message should appear, and clicking “Delete” will wipe out your saved clips.”