Your Phone Is Listening and It's Not Paranoia (2018) | Hacker News Reader

Your Phone Is Listening and It's Not Paranoia (2018)

vice.com

14 pointstlrobinson10 comments

I hate these anecdotal stories and how they're used as absolute evidence that something like this is happening. Google would never allow the data from their back end recording to be seen by other apps, for the same reason that this article says Facebook keeps the data they collect close hold: it gives away their exclusivity.

Seriously. You met with a friend at some place where you probably sat around and talked for a long time together. Is it unreasonable that FB saw the two of you were together at a place where you might reminisce about previous trips together.

As for saying specific phrases near your phone, this sounds like Baader-Meinhof to me. The writer probably spent a lot longer on FB than usual, which led to more ads (and an expanding category of ads to keep from repeating the same ones over and over), leading to him finally seeing what he was hoping to see.

There's just no way that either Google or Apple are storing the voice recording data anywhere where other apps could see it. Further, I don't think either are maintaining recordings long enough to make sense of a conversation or even full sentence. In Google's case, they've talked ad nauseum about how it's just a voice signature they listen for and they never maintain non-triggered audio (even device-side).

The thing I always come back to with these claims is battery life and/or data usage. If your phone was always listening to you, it would need to either be always streaming the audio data back to home base, in which case you’d notice the data usage, or it would be doing on-device voice recognition, which would cause noticeable battery drain. Both cases would make it very easy to detect and this question would already be answered definitively.

"If your phone was always listening to you, it would need to either be always streaming the audio data back to home base"

Not necessarily. It could also send the data intermittently, in batches.

Even then, of course, there would be data usage. But you might not notice it. It might be hidden as part of the data use of various Google apps.

For instance, on my phone the "Google Play Store", "Android OS", "Google Services" and "Google" apps regularly send or receive many megabytes of data. What data they're sending is completely opaque to me. If it was recorded voice data, I'd never know.

You could test your local drain hypothesis pretty easily. Turn on voice activation of e.g. Google on an Android. This means your device is now listening to and analyzing every word to see if it ever recognizes 'Okay Google'. If your theory is correct you'd notice a substantial battery drain compared to having this functionality disabled. I'm going to guess you will notice basically nothing. The nice thing about neural network driven systems is that while they are very resource intensive to build, once built they are (generally) extremely resource friendly. And training a network to recognize one of a wide array of terms would be no more difficult than training one to recognize Okay Google. And the best thing is that false hits don't really matter that much, if you assume the only motivation is advertising.

Though, for what it's worth, my bet would be on exfiltration of the data instead of local analysis. Put the shoe on the other foot. Imagine your goal is to exfiltrate data from a device in a covert fashion. You can come up with countless clever ways starting at the most low level with extreme compression. All that matters is that what is said can be recovered - it doesn't need to sound pleasant, or even intelligible as long as another system can reconstruct the speech (even if at great computational expense) from what you send. And from there you can come up with all sorts of clever ideas such as only sending the data during other normal transmissions.

Is the claim that these apps have special deals and APIs with Google and Apple to access this data?

There are certainly no public APIs to get these recordings (if they exist). The claims in the article seem pretty audacious and are presented with no evidence.

I read recently that some things are explained in part by a data sharing agreement between amazon and Facebook

https://gizmodo.com/amazon-and-facebook-reportedly-had-a-sec...

2018

and past comments. https://news.ycombinator.com/item?id=17255265

note that the submission was flagged.

Related: Am I wrong that Facebook's apps cannot access random conversation without having been granted microphone access on Android?

I don't even have the app installed, but I see similarly relevant stuff occasionally when I browse to FB, which tells me that there gathering enough knowledge Annie me from a plethora of other sources.

Reply on news.ycombinator.com