[intellent]

I really like the sound quality of Amazon’s newest Echo Dot generation compared to it size and price.

However, in my experience Alexa is still pretty stupid. The only things it does well, is to play music, switch lights and read Wikipedia articles.

The one thing I'm scared of is Alexa sending everything it hears to Amazon, and not only things I said after the trigger word. I know for a fact, Alexa constantly records even today, because it can filter "trigger word in the middle of a sentence" scenarios, when no actual request is intended. I really do hope, that only actual requests (or false-positives) are currently sent to the AWS cloud.

If that won’t be the case anymore in the future, I’ll be out.

[SavageBeast]

I spent some time taking apart how Alexa must work myself. My thinking on this is guided by the fact that there are only a handful of Wake Words available. Given that there are only 4, I'm thinking the ability to recognize those Wake Words is in firmware on the device (such tech was commercially available back in 1995 when I played with a demo piece of hardware that did just that). Decoding a wake word from a stream of audio locally on the device without making a persistent recording of it or sending a recording off-device is fine with me.

However the more complex parsing of what ever comes after the Wake Word needs to be sent off as an audio stream to some cloud operation for parsing where more horsepower is available. This also seems quite reasonable to me. Sure, it could be done on the device but this would mean a hotter CPU on the device and a generally more expensive Alexa. Certainly no more $29 Echo Dots.

Is Alexa listening all the time? Hell yes it is! Is Alexa recording everything you say, everything you watch on TV, every time the dog barks and sending it back to Amazon for archival? Its possible but consider the economics of storing that much data as audio. Storage is cheap but thats quite a burden. Is AMZN keeping a parsed version of everything it hears from your device and dumping the raw audio version? Its technically possible but again, consider the economics of such a thing.

Not to mention, how many Alexas are going to be sold after someone here posts smoking gun proof Alexa is constantly or even at some batched interval sending off large chunks of data which are in fact full voice recordings?

I'm not going to waste my time on this experiment but anybody who is good with Wireshark could perform an analysis to quickly disprove this whole notion.

[ChicagoBoy11]

You have to be careful not to conflate two things. It is immediately obvious that Alexa "is always listening" -- after all, how could it know you are about to utter the wake word? So, in a sense, the device is listening to you at all times.

That, however, is separate from a concern that the device is streaming your data back at all times. The implementation could easily locally listen for the wake word, record your utterance, and only stream/process that part of it. It can, in a sense, listen at all times but only tell Amazon what you actually want it to tell Amazon. Which, AFAIK, is how the device works. That doesn't, of course, mean that it works flawlessly or that things are captured which you don't intend it to capture, which is a fact that has also been covered extensively by the media reporting on Alexa.

[openasocket]

Disclaimer: AWS employee, not remotely related to any Alexa work

From the public record you can actually be pretty certain that Alexa doesn't send anything back to Amazon other than actual requests (and false positives). There was a murder trial a few years ago where an Alexa may have grabbed some audio of the murder. A judge ordered Amazon to turn over anything they may have, and Amazon said they didn't receive any requests during that time period, and they don't get any recordings other than those requests. So Amazon has effectively stated, under threat of criminal prosecution, that they don't record anything other than those requests.