Point 1: Keyword triggers and is actually said are miles apart, as shown by the Alexa creepy laugh episode. There is no practical way of knowing when Alexa has heard a keyword and recorded content around it. Not to mention the Google play incident where a poor wiring job made the devices think the "enable assistant" button was constantly being pressed.

Point 2: My phone, an iPhone, offers a way to turn off the keyword-listening functionality - and presents this switch during initial setup. It also offers a way to turn off all tracking functionality. Apple has no financial gain from lying about this.

Point 3: This comes back to point 1. If you don't know what and when the assistant is listening to, you have no idea what phrases are being collected and shipped off. Also, there are plenty of studies floating about the internet about how "anonymized" search engine queries have been collected, studied, and in many cases paint very personal portraits of the people doing searches.

> There is no practical way of knowing when Alexa has heard a keyword and recorded content around it.

I know this isn't the kind of certainty you're looking for, but Google Home and Alexa both allow you to enable audio cues for when they start and stop recording. I find this helpful from an accessibility perspective, but the privacy-conscious may also appreciate the indicator.

They also have visual cues enabled by default. The light turns on when it is listening. You quickly get an idea of when it is recording and when it is not. And it rarely falsely triggers.

Point 2: Although you’re right about the application processor side of your phone, you cannot disable a significant amount of tracking functionality built into the baseband/modem. Your cell carrier or ISP can track you the same way that Google could on an Android device. On the other hand, iPhone basebands usually don’t have DMA, while other phone manufacturers allow the modem to access application processor RAM.

The iPhone has a lot less tracking built in and has much better privacy controls than Android with Google Play Services, but there are still some avenues for different companies to track you and to have a lot of control over your device.

Cell carriers can track iPhones as well through triangulation, etc. There’s not much you can do to prevent this.

You can simply turn on airplane mode?

Re: point 3, I was reading Kevin Murphy's book on Machine Learning recently and in it he says that roughly 20% of Google search queries that come in everyday have never been seen before. Reminds me of that report about Russian election interference campaigns that included the tidbit about Google being able to pinpoint the exact search queries that members of some GRU unit (Fancy Bear, maybe? idk) made.

> There is no practical way of knowing when Alexa has heard a keyword and recorded content around it.

This is incorrect. If you want to know what your device has heard and processed (or failed to understand), you can look at the Alexa app on your phone. It will tell you what it thought you asked for and what it told you or played for you, and you can replay the recording of your voice request.

You can opt out of basically every Google login/tracking in Android, except the Play store. I know because I'm logged out of everything and I disable location services. Google Maps keeps working.

What I don't understand is why they have to store voice recordings instead of deleting them once they understand the vocal command. There could be a number of reasons and I don't like any of them.

>What I don't understand is why they have to store voice recordings instead of deleting them once they understand the vocal command.

They don't have to. You can disable it at https://myactivity.google.com

try disabling Google Maps location access permission. For me it triggers an endless loop requesting Location permission, rendering the app unusable.

Google's always mining data behind the scenes and gapps are constantly phoning home, your "opt-out" preferences are meaningless.

This is one of those massive privacy violations that I cannot believe has been allowed to exist as long as it has without people making a concerted effort against it. I remember way back in the early days of Android when it was excused for reasons of allowing Google's live traffic mapping feature to work.

That excuse made sense in a time where public trust in these companies was justifiably high but it's turned into a situation where the major tech giants have the power to become a modern day Stasi on steroids (and increasingly show their willingness to play that role). We had to place enormous trust in these company's ethics (mostly on a subconscious level as they all used to go to great lengths to afford us the ability to implicitly trust them) in order for the smartphone/mobile revolution to occur. Now that these devices are a necessity in modern society, they've all dropped the pretenses that allowed us to trust anything to do with the information they're collecting.

We're living in a cyberpunk dystopia and I don't see an easy way out.

Fortunately OSMand works reliably (and offline).

It's to train their speech recognition and intent recognition systems. They probably also use the data to prioritise which queries they add support for.

>except the Play store

Workaround:

1)Install F-Droid (f-droid.org) 2) Install Yalp Store from F-Droid

Even if you do this, on a standard Android device, there is still Google Play Services installed, which has every permission available on your device. You have to install a fresh custom ROM like LineageOS without GApps to avoid this.

Good point. You should install LineageOS first, you can then select the GApps bundle size you need or try and replace them with FOSS alternatives. For me, Yalp Store has made using LineageOS + minimal, permission sandboxed google services a bearable experience.

> What I don't understand is why they have to store voice recordings instead of deleting them once they understand the vocal command. There could be a number of reasons and I don't like any of them.

Not that I have any insider information but a while back someone made a post on reddit about what baffles them about their work and they had a job where they basically listened to very short voice clips and saw something on the screen and they had to decide whether it was the right word on the screen or not.

There are tons of phone systems that use speech recognition and warn you that you are being recorded for quality assurance purposes. I would assume those recordings come from that type of system. Google and Amazon have been clear that humans will not be listening to the recordings.

Are you trying to say these home assistants are basically mechanical Turks?

Most likely the human verification is not in real time but for quality improvements after the facts.

In case you weren’t joking.

> Literally every article about home assistants brings out a number of people who continue to say how they can't understand how people could have this in their homes.

It feels weird to feel out of touch. I’ve been a computer/technology enthusiast my whole life, and used to proudly early-adopt all sorts of crazy gadgetry. But the older I get, the fewer things coming out of Silicon Valley appeal to me, and the less I understand the use cases. I just have to shake my head and accept that they are somehow useful to a lot of people and that the idea of a cool useful product is changing.

"Alexa, play [almost any recorded music ever]" is pretty cool.*

* You have to have Amazon Music Unlimited which costs from $4.99 to $9.99 a month.

I would love to have an always-on always-listening home assistant which did not ship my audio stream off to a cloud I don't control.

The fact that all the current listening assistants do ship the audio stream to the cloud has nothing to do with CPU power and everything to do with a desire to collect as much personal data as possible, and as far as I'm concerned they can pound sand.

I get 1) and 3), but I don’t buy 2). Your phone works with Siri/any other assistant off. The connected speakers don’t. So the comparison doesn’t hold in my opinion.

As an example for #2, when I got an Android somehow I was either automatically opted in or accidentally opted into location sharing. Two years later I found that Google Maps could show me every location I had been to for the last 2 years. I could review, day by day, everywhere I had gone, what time I went, how I drove there and where I went to next.

You can just mute the connected speakers and it's the same as having Siri off. Or get one of the ones that requires you to push a button (the Amazon Tap).

Many phones work the same way, the phone is always listening for you to say ok siri.

Can't you turn of this feature on iOS? I don't have Google assistant on on my device.

Yes, you can. It’s in Settings under “Siri”.

1) It's always listening for a keyword, and then it starts recording. It's a voice interface for a computer.

Show me the source code!

...all of it, including what's running the entire cloud behind it! Until then, fuck no, no hot mics in my house. And actually then, still no hot mics unless I've plugged my Les Paul into my amp, in which case I switched them on.

It's rock and roll. You don't get it, but somebody does.

1) Nobody can prove that. It's always listening and you have to assume always recording and transmitting.

2) My phone is a $12 throw away that speaks voice and sms (text).

3) Loose lips sink ships.

If we are playing the "you can't prove that" game, then you also can't prove that your $12 throw away phone isn't always recording, or that you new kitchen countertop doesn't have a recording device embedded in it, or that the maker of whatever device you are using to type that comment isn't breaking a plethora of laws to record everything you type or say or do around that device at any time.

It's reasonable to embed motive into these suspicions. My confidence in iPhone not listening to me is linked to the fact that Apple's ad network is a fraction of the size of Amazon's and Google's, and that to the best of our knowledge, it doesn't require extensive access to personal data to run effectively. Same with GE appliances and countertop makers. One of the most reliable ways for me to believe a company will not harvest my data surreptitiously, now or with TOS updates in the future, is for me to be convinced it's not in their best interest to do so.

True but...

1- Apple might not be interested in listening to its users but I'm sure other businesses are. Apple users are loyal customers who can pay twice for a piece of technology and some of them will happily spend a night in line to be able to do that. Most advertisers would pay big money to Apple to know who those people are and what they want; that information is like gold.

2- unfortunately advertisers aren't the only entity potentially interested in spying people.

> Most advertisers would pay big money to Apple to know who those people are and what they want; that information is like gold.

I think its laughable that this money would be tempting to Apple when they have a business that almost sees customers buying 700-1000 dollar plus devices on an annual or bi-annual subscription, especially when the privacy stance is one of the core pillars of the marketing message.

There is likely no commercially viable business model Apple could adopt to sell that data that would be remotely worth risking a 60-70 billion dollar a quarter in revenue money printing machine. That information is practically worthless in the context of iPhone revenue.

Should iPhone sales start to falter, this is absolutely something I would consider a concern, but the iPhone has a long way to fall.

At least you can pull the battery out of most "burner" tier dumbphones.

My battery falls out if I stare at the phone for too long. This works on goats, too.

When my throw away phone is transmitting, it gets warm. I also set it next to an amplifier and I can actually hear it bleed into the amp when it chats with the cell site. I can even tell a couple of seconds ahead of time when I am getting a text or a phone call.

And all a hypothetical "bug" needs to do is listen for some key phrases, and send a single byte or a few bytes of data across the wire, even through side channels.

I can imagine a system that throws an extra byte or 2 on those "chats" with the cell tower, or even abuses the timing that it decides to reach out and check the tower as the method of communication.

The actual voice recording and processing tech is small enough that it could be thrown right alongside the actual phone, for pretty cheap, and could run entirely separate from the phone itself, maybe even having its own battery!

Yeah, but you can also do the same with the Alexa device by watching it transmit data over the wire. If you think watching the transmissions is enough to prove your cell phone isn't recording, then it should be enough to prove the Alexa isn't recording.

Or not be sufficient in both cases.

> When my throw away phone is transmitting, it gets warm

Maybe that's just what they want you to think so you don't dig any deeper.

I vaguely remember rumors (or maybe it was used in a few court cases by the FBI), where carriers would turn on the microphones of 2G dumb phones remotely and silently. So your dumb phone probably has the same issue.

Carriers also keep records of where subscriber devices have been, so even if your phone doesn't have GPS, the carrier has cell tower triangulation logs of the device too.

ISPs such as comcast also save your browsing history to run their advertising networks, so that is also recorded somewhere too. And if/when net neutrality goes away completely, they might just sell that info too:

https://arstechnica.com/tech-policy/2017/03/comcast-we-wont-...

Whats even worse about ISPs doing it is how you usually have no other choices and it's your complete traffic stream!

Steve Gibson did a wireshark investigation of an Alexa, it is NOT always transmitting.

Your assertion that "nobody can prove that" is rather absurd.

there's nothing stopping from Alexa recording content locally and transmitting it with a traditional request once activated.

Also, there's no reason for Amazon to want a constant stream of Alexa data hitting their cloud services. The goal is to mine and extract valuable data about users to build profiles, ideally you'd want to do that on-device and send up the most valuable bits.

These are also capabilities they can silently introduce at anytime for any device.

Steve Gibsons "analysis" means nothing.

Except, you know, memory - of which these devices don't have a ton.

You can store around 1000 hours of recorded speech in a gigabyte of flash memory that costs $0.25. For most people, that could hold their entire conversational activity for a year. Storage and off-line upload (perhaps during a software update when packets are flying around anyway) isn't difficult.

They can not silently introduce new hardware to these devices. You should look into what kind of hardware these devices actually contain. Try to set up an intelligent, personalized data miner with those constraints.

Steve was not thinking this through. I can batch compress chunks of audio and lazy upload them when I think nobody is around. A listening device could probably even guess when you are not going to be using your internet connection.

You can filter the captured traffic for anything coming from Alexa's IP or mac, or transmitting to any amazon-owned server.

Sure, but you wont know what it is or why it is.

> Nobody can prove that.

You guys are way to paranoid. They're a public company and they are constantly under audit. They have explicitly stated in their white papers that the device has two computers, one that just listens for the "Alexa" keyword that activates the other one for signal processing.

Do you really think they would lie on those white papers? The government would be so far up their ass with fines it would be ridiculous. Not to mention people have actually done tear downs of the device and you can see the two computers and monitor the power input to each one as it runs. The "dumb" computer with minimal memory would be having to store hours of speech data until you say the keyword so it can transmit to the other computer.

Audits of controls are about ensuring the longevity of the business. They are entirely orthogonal to ethics or business models.

And yes, I am proudly paranoid. History and experience have confirmed the need for my paranoia for many decades.

> Audits of controls are about ensuring the longevity of the business.

You must not have suffered a SOX audit before. Part of the audit is technical.

Also, the SEC greatly frowns upon lying about your business.

I have suffered every kind of audit, many times. I know of many technical facets of every business I worked for that would never have been a part of any of those audits.

Do you know of any published Amazon audits that describe in detail what specifically around Alexa is audited, if anything?

Yes, but is that behavior changeable with only a software update?

It's not paranoia if you're right.

Actually, quite the contrary has been proven to point 1: both for Alexa (the spooky laugh episode) and Google Play (the "always pressed" button episode).

I almost think that the burden of proof that things aren't being sent is a valid expectation in this point. And that proof needs to be re-confirmed with each firmware update to ensure a bug hasn't caused the behavior to change.

Those aren't examples of the contrary, those are examples of in-precise recognition. That's miles away from being the same thing as "always recording".

And your SMS is logged as well as cell towers to which you connect. Is the “phone company” more secure than any other? If you really are someone in need of a tin foil hat, you should ditch the phone completely and communicate solely with dead drops and one-time pads.

SMS is actually clear text over SS7. I would not send anything I care about over it. It is also trivial to spoof.

All you need is 1 SS7 link and you can be any phone, kick anyone off a cell site, override call congestion and bump people off, change the display on your phone to any text, too many things to name.

Fun to watch the venerable "tin foil hat" lose it's luster.

> Nobody can prove that. It's always listening and you have to assume always recording and transmitting

Uh... packet analyzers, IP logs, bandwidth monitoring tools etc...

If instead of transmitting occasional bursts of data when you say the wake word it's been sending back tens of megabytes an hour, or tens to hundreds of megabytes a day, you can go "hey, it's probably transmitting everything it hears".

There are people that religiously watch what these devices are sending back just to try and catch one of the companies doing something fishy. There are people that religiously watch their network traffic period looking for something nefarious.

These devices aren't like your phone, where simply having it on can have all sorts of services/apps using data in the background or you open a webpage and tens or even hundreds of different servers are contacted and data pulled/pushed to/from them. It's a device that should be periodically checking for an update and sending very small audio files back in nearly-real time. If they were recording regularly, or especially all the time, and transmitting it back to a server it would stand out like a sore thumb.

Never mind the fact that keeping it secret, and not having a whistle blower, would be damn impossible. Someone's going to mention it to a loved one or friend, or flat out go online "I discovered my Amazon/Google device is actively listening and recording and sending to these IPs in these intervals" when they just know it as fact and didn't have to 'discover' it.

Which nobody can do. It is encrypted and unless the device accept invalid certs, they can queue up, compress, batch upload any data they want. They control the security, not the end-luser.

We know how much storage is on the device - we could tell thru network monitoring if they were actually doing this - there is just enough smarts in the device to only turn on the recordings when the wake word is spoken - thats not being said a malicious player wouldnt hack it, and insert their own code to listen all the time - but based on the evidence, I don't think amazon is that player.

Encrypted or not, an audio stream should not be difficult to identify.

If they batch compress and upload, it wont be a stream. It could be aes-256 lzma2 compressed chunks. Using the right codec, they could likely upload a hour of audio in a few hundred kbytes. I have seen some demos of highly bw optimized codecs.

Here are some staring points. [1] I would go with Opus. My voice chat server uses that.

[1] - https://stackoverflow.com/questions/167533/best-voice-compre...

> Using the right codec, they could likely upload a hour of audio in a few hundred kbytes.

But at the same time, you're convinced that you would notice if your phone was slipping out these kilobytes as part of its usual communication?

It seems to me that you've just arbitrarily picked a level of paranoia where you've decided that this can definitely happen with the Alexa, but won't happen with your phone. Your proud paranoia isn't applied consistently.

I am not convinced of anything.

The risk of the throw away device uploading my audio is significantly lower than a device powered in my home, with dedicated fast internet.

What if they convert speech to text like the American intel agencies do as proved by the Snowden and Vault 7 leaks?

That's possible but that's quite a bit of processing for actionable speech to text which means you have a chance of the device acting sluggish when you query it or you have to give the device way more processing power than it needs which increases cost and from my understanding with devices like Amazon's, they're already being sold relatively near cost because they are intended to get you to spend more money inside the Amazon ecosystem.

Can you MITM SSL the conversations to Amazon? It has memory. There is no requirement that it transmits in real time.

>The utility of being able to interact with a computer via my voice wildly surpasses the risk.

How?

Risks:

Willingly putting a proprietary device designed to listen to you and watch you, created by a company that exists only to maximize it's shareholders profits by either selling you things, advertising to you or collecting and selling data about you in order to do those first two things better, into your home to constantly watch and listen to everything that goes on.

Utility:

Control a computer by talking to it...

For me at least....that risk far outweighs whatever utility i migbt get by being able to speak to a computer and honestly, if i really needed it that bad i would just set up my computer with voice recognition and control software. It's existed for a while now. I remember playing with it when i was a kid...it sucked...but that was more than 15 years ago.

The only real novelty of alexa is that it'a a standalone box that answers you back.