Hacker News Reader

Comment by bryanlarsen | Hacker News Reader

I understand the concern, but what makes a smart speaker any different from a smart phone? They also have a microphone that can eavesdrop on conversations, are harder to monitor for unexpected traffic and also provide location information.

How many people that refuse to have a smart speaker also refuse to have a smart phone?

View on HN

There’s a clear difference in power limitations between phones and speakers. Phones are almost always on battery, so what they can gather is limited. Speakers always have full power, so keeping the electronics active all the time has no downside.

There's also a clear difference in hardware limitations. Phones are orders of magnitude more powerful than smart speakers.

Phones only listen when you use them. Speakers listen all the time. Phone calls are usually not transcribed where speaker recordings are, and can be searched back in time as well. Phone calls are not recorded, though metadata is.

So basically a “smart speaker” is like having 24/7/365 surveillance in your home.

Both the phones and smart speakers start listening on wake words. I haven't seen any thing to suggest the smart speakers are constantly listening and sending everything to Amazon/Google/Apple. But both clearly could be constantly recording, not just when you're making a phone call. They're the same threat model, in my opinion. Plus the phone goes with you to places the speaker doesn't.

If it's not listening, then how does it hear the wake word? Hint, it is. It just doesn't start parsing the request until it hears the wake word.

Unless you explicitly turn it off.

Maybe a Google engineer can correct me but I'm pretty sure there is no meaningful distinction to be drawn between the Google Assistant running on the Home devices and the Google Assistant running on phones.

Having a "smart phone" is also like having 24/7/365 surveillance in your home.

Voice assistants on phones can be set to only trigger on manual interaction, and they can be turned off entirely.

Having an always on voice assistant is like having 24/7/365 surveillance in your home, regardless of what device it's on. But having a smart phone does not require you to have an always on voice assistant.

But how is the action (e.g. an over-the-air patch/hack) needed to secretly convert an Alexa to always-listen-and-transmit-home different than it would be for a smartphone?

I'm not worried about an over-the-air patch or hack[0], I'm worried about the existing problem Alexa has with false positive wakewords. Voice recognition isn't reliable enough yet to use as a security feature.

https://www.vox.com/recode/2020/2/21/21032140/alexa-amazon-g...

https://www.independent.co.uk/life-style/gadgets-and-tech/ne...

[0]: I mean, I am, but not in this specific case. I don't think smart speakers represent a particularly unique risk where hacking is concerned (Nest devices excluded).

The entire point is that you're trusting the manufacturer that the device implement the behavior the manufacturer says it does. If you have a smart speaker, you're trusting they don't upload all speech.

With a cell phone, it is actually tougher, because you're also trusting the manufacturer that their security model is such that 3rd party software can't violate the privacy policy you're expecting. Many, many free phone apps today collect data you don't want them to.

> If you have a smart speaker, you're trusting they don't upload all speech.

Well, you shouldn't be doing that because we've known for a while that smart speakers have a false-positive problem.

That's the advantage of a physical gesture; the AI won't get it wrong. A physical action like holding down your home button is much less likely to trigger on accident.

Not to mention, you can also full-on disable the voice assistant on your smartphone, in which case, there's practically zero risk of a false positive.

> With a cell phone, it is actually tougher, because you're also trusting the manufacturer that their security model is such that 3rd party software can't violate the privacy policy you're expecting.

Stealing my contacts is bad, getting access to my microphone is worse.

If the argument is, "my phone might get malware, so having an always-on-microphone is exactly the same", that just seems really fatalistic to me. These are different risks that belong in different categories.

Sure, but without the phone OS source code, you can't know whether "disabling the smart assistant" just displays "disabled" and then keeps recording. Without a hardware switch, it could just lie to you.

The risk of my phone OS being theoretically compromised to spy on me is different than the risk posed by a device that I know is listening to me, and that I know regularly misinterprets wakewords.

It's fatalistic to say, "I can't see the source code, so I might as well install another microphone in my bedroom."

Yep, you're right on this front. It's not a binary thing. That is, you are concerned at inadvertent data leakage that could be used against you.

For what it's worth, there's a hardware switch on Echos by the way.

Are there any custom ROMs for home assistant microphones?

I don't believe so.

I think the difference is that speakers are meant to be always-listening. This is less possible on a phone where it would be detectable.

Phones are equally always-listening.

They respond to "Hey Siri" or "Hey Google".

There is absolutely zero difference between a phone and a smart speaker in this regard.

I don't know why you think anything would be more detectable on a phone.

There's a ton of difference. My phone is not always listening -- its voice controls aren't enabled, and even if they were enabled they would be hidden behind a button press.

Sure, somebody could hack my phone and take over the microphone, but that's a very different threat model than "this AI might misinterpret a random sound as a command and start uploading my conversations to the phone."

To the extent that normal people leave their phones in always-listening mode, we should be educating them about the privacy implications of that as well.

You can mute the mic on a smart speaker too. Doesn't change the fact that the vast majority of people have them always on, just as they do with their phones.

It's far, far easier in practice to use an unambiguous gesture like a button press on a phone than it is to use an unambiguous gesture with a smart speaker.

Most normal people could get by with a voice assistant on their phone that required manual prompting, but a voice assistant like Alexa that required you to walk across the room and push a button to use it would be useless -- at that point carrying your phone in your pocket would be a better UX experience.

How the products are designed and implemented in the real world matters: phones are designed to be in-reach of people, smart speakers are not. You can't just stick a mute button on a smart speaker and say it's the same UX as phones now.

> How many people that refuse to have a smart speaker also refuse to have a smart phone?

If somebody is avoiding a smart speaker out of privacy concerns, but they have an always-on voice assistant on their phone, the correct response to their concerns is not to tell them to stop worrying about smart speakers. The correct response is to educate them about their phone settings so they can change them and exercise more agency over their lives.

iPhone: if I say 'Hey Siri', I see an onscreen animation indicating it's listening and ready for my next command. Alexa: if I say 'Alexa', I see the ring lights animate indicating it's listening and ready for my next command.

Putting Amazon/Apple's privacy and security implementations aside, how are these different? Aren't they both 'always listening'?

You can set your phone to require a physical action before it will start listening. On Android, I think it's a long press on the home button, or it was the last time I was looking at the controls. You can also disable both 'Hey Siri' and 'Hey Google' on your phone entirely.

If you are using a voice assistant on your phone, and if you have it set up to always listen, then it's almost the same as a smart speaker (minor quibbles about the positioning and quality of the microphone aside). But those are two pretty big ifs.

Usually this argument gets pulled out to shut down people who have concerns about smart speakers in general, saying that any effort to make smart speakers more private or avoid them is pointless because of course their phone is always listening to them. That's not necessarily true. If someone wants to get rid of their smart speaker, owning a phone doesn't immediately make them a hypocrite. They can keep their phone and still have better privacy.

you are assuming the physical action before listening will always remain a faithful setting. If the NSA wanted Apple to play ball one day, that setting would be merely cosmetic.

I'm not assuming anything, I just understand that the risk of the NSA forcing Apple to install malware on my phone is lower than the risk of my local police office following an already legal process to get data that I know Amazon is inadvertently collecting from people today.

You're correlating risks that aren't related to each other; these are two different devices with different threat models. If the NSA wants to bug you, it will bug your house. That doesn't mean you should bug it for them.

Power usage and data usage are much much obvious on a mobile device. But not on a wall-powered home-wifi smart speaker.

The speakers are always-listening in exactly the same way that most smartphones are always-listening: They are always passively listening for the wake-word. Without the wake-word (or at least something that sounds like it) they do not record or transmit audio data. All the major smart speaker brands clearly indicate when they have been activated with lights and/or sounds. For the truly paranoid there is generally a switch to disconnect the microphone.

Reply on news.ycombinator.com