Hacker News Reader

Comment by zie | Hacker News Reader

I have no doubt they are able to hear anything happening in the room they happen to be installed in. But I doubt they listen for anything but their keyword, and I seriously doubt they listen, record and send off to their overlords what they hear on a massive basis.

However, they clearly CAN, and while they almost certainly don't on any massive scale, if they by their overlord or a security hack, did on a targeted basis listen, record and send conversations on, that would be very possible, and very harmful for whoever was targeted.

I can't imagine the NSA and other covert organizations not drooling at the chance to do just that. I'm also fairly certain those devices are not secure as by and large the tech industry has a complete failure rate at making anything computing related secure.

View on HN

The FBI has been doing this for 15 years with cars and OnStar. They will do this with smart speakers once enough people have them.

https://www.forbes.com/sites/thomasbrewster/2017/01/15/polic...

2006: The FBI admits for the first time to turning a dumb phone into a microphone to listen in on a some New York mafia guys conversations at a restaurant in the early days even before smartphones made it even easier to do so. Which they publicly used to convict them (meaning it was probably an old tactic by then):

> The FBI appears to have begun using a novel form of electronic surveillance in criminal investigations: remotely activating a mobile phone's microphone and using it to eavesdrop on nearby conversations.

https://www.cnet.com/news/fbi-taps-cell-phone-mic-as-eavesdr...

And we can't reign it in cause terrorism!

>However, they clearly CAN, and while they almost certainly don't on any massive scale, if they by their overlord or a security hack, did on a targeted basis listen, record and send conversations on, that would be very possible, and very harmful for whoever was targeted.

Besides the fact that they have better antennas, this is true for pretty much any device with a microphone and a processor with audio processing capability, no?

My laptop has a microphone and video. Surely Microsoft could just start recording with the with right windows update?

My Avaya VOIP phone has speaker phone, could do the same.

> My laptop has a microphone and video. Surely Microsoft could just start recording with the with right windows update?

It's a commonly mentioned issue, and a somewhat regular occurrence with various malware or "security solutions".

That's why a number of laptops have either visual indicators or shutters (though mostly for the cameras, sadly hot mics are rarely a consideration), and a rare few have physical disconnection switches (again mostly for cameras).

I saw one gizmo for mic the other day

https://shop.googlemerchandisestore.com/Google+Redesign/Acce...

haven't tried it though, anyone has any experience? Does it really muffle the sound enough?

I think the idea there is that it physically disconnects the mic, since the laptop instead tries to use the headphone mic (which in this case isn't connected). Not sure if it would actually work, it would be easy to 3d-print one to test.

If you connect a plug to the audio jack, sure, the laptop probably defaults to using that jack for input and output. But you can still choose to use the built-in mic, right? So it's not physically disconnecting anything, just triggering a software disconnect.

AFAIK you're correct. Though it's possible most spying software just uses the default mic and doesn't have the capabilities to switch the input device?

Not a Mac user, and the last time I played with sound stuff in nix was almost 15 years ago...

But I feel like switching input (or even collecting multiple sound inputs) on a nix OS is not much more convoluted than dealing with the sound stack in general. But that's probably also related to how the driver itself interfaces with the audio device...

Macbooks since 2018 physically disable microphones and webcams when the lid is closed: https://www.businessinsider.com/apple-macbook-air-and-2018-m...

How does a chip "physically disconnects" a device?

Presumably power is disconnected by the T2 chip. Different from mechanical disconnect via a physical switch, but equivalently effective if you trust the T2 chip. (And if you don’t ... well, you can’t use the MacBook securely at all.)

You had better not trust the T2 chip, because it is vulnerable to the checkm8 exploit and the checkra1n folks have already demonstrated total compromise. The encryption functionality isn't affected if you have FileVault on, because your password is not stored anywhere on the device, but everything else, from basic SMC functions like mic/cam/fans/touchbar to secure boot to verifying the microcode and ME firmware before loading are totally useless now.

AFAIK, the T2 is always powered on even when the main CPU is off, so this could have ultra-long-term persistence.

Reference?

https://www.idownloadblog.com/2020/03/10/luca-todesco-teases...

Because the mic and camera runs through said chip.

But the lid is open for several hours a day while I'm working!

That would be a big reason to get a MacBook for the first time ever.

Perhaps within the hazy, inextricable mists of jack, pulse and alsa, there's an answer. I remember feeling qualified for a job at NSA after getting midi and my mic to work (effectively) on my Arch (Mint, Ubuntu, Debian, etc) laptops years ago. It was a feat. I can see:

Hacker1: we've got root.

Hacker2: yeah, but how the fuck do we get audio to work?

Hacker1: modify grub to default to windows partition and reboot, or maybe install Virtual Box and run windows. Probably not possible otherwise.

Hacker2: No, waaay too slow, they're on an ATT gateway. Find a softer target.

> Surely Microsoft could just start recording with the with right windows update?

Yes, that's why automatic updates was seen with a lot of skepticism at first. And the reason automatic updates are winning is because everything is so full of holes that disabling automatic updates is even less secure than enabling it.

Surely Microsoft could just start recording with the with right windows update?

Yes, they could. I get some scepticism when I say my businesses have declined to move to Windows 10, mostly because we don't trust the security and privacy aspects. However, we deal with personal data, and we also deal with various types of information that are protected by statute and/or contract. It would be very obviously against the spirit and quite possibly against the letter of several different laws for us to knowingly store and use that kind of data on a system that sends information we can't control up to the mothership and/or that could be updated without us agreeing to it. Indeed, there are multiple ongoing investigations into Windows 10 because of exactly those kinds of concerns across the EU right now.

It's a good thing that non-technical people who do take privacy seriously, such as lawyers, are starting to notice the glaring problem with these modern technologies and advise against using them. With a bit of luck, we'll then get statutory regulation requiring disclosure in advance of exactly what these devices are doing and the security and privacy implications including when they don't work properly or if they get compromised. Again, there had been much discussion in Europe recently, at least up until the world was more concerned with another kind of virus over the past few weeks, of requiring much stronger standards for security in IoT devices, and those sorts of laws really can't arrive soon enough. Mandating controls to physically isolate all cameras, microphones, transmitters and receivers at the hardware level might not be a bad idea, either.

There were stories earlier about Alexa employees and contractors hearing all kinds of unintended conversations.

https://www.latimes.com/business/la-fi-tn-amazon-alexa-echo-...

It should be noted that the device is not intended to record conversations absent the trigger word, however there can be false positives:

> According to Amazon’s website, no audio is stored unless Echo detects the wake word or is activated by pressing a button. But sometimes Alexa appears to begin recording without any prompt at all, and the audio files start with a blaring television or unintelligible noise.

Well, from a suspicious mindset that could be more or less sincere...

It doesn't even matter how sincere they are. False negatives hurt the perception of the product, because then people need to repeat the key word. False positives do not hurt the perception of the product, because the recording occurs silently. Their incentives are aligned against privacy, and so privacy will suffer.

False positives does hurt the perception, because the device will usually do something or utter a response unless it fails to detect something intelligible or writes the request off as not intended for it (having two devices at opposite ends of a large open space living area can sometimes get frustrating, but looking at the history, you see how often they correctly determine which device should respond)

An occasional false positive is funny, but it gets old pretty quickly.

That's not true. Amazon can have a significantly more accurate classifier running on their backend that processes the transmitted audio. And this classifier can be more accurate due to both more computation and it would have access to the entire data stream, not just the initialization words. You would never know that your data was sent to Amazon and then ignored.

They can, and they do. You can see that in your history because you can listen to the audio of things it decided subsequently was not intended for it.

But unless you have multiple devices, those make up a relatively small proportion of requests, and hence most of the time false positives is something they have a strong incentive to stop because it causes negative user experiences.

[If you have multiple devices it will happen "all the time" when a request is genuinely meant for a Alexa device but more than one device hears the request, but then it has no UI effect - one device will usually answer, and the other(s) will stand down; this is by no means perfect, but it works reasonably well]

Obviously, any device which is mic'd and connected is a risk. And devices like the echo or whatever google and apple sell have a permanent hot mic' since they keep trying to match the keyword so a "live" indicator would be of very limited use.

Unless the trigger-matching was performed by some sort of tamper-resistant secure module which would also trigger the indicator and not feed any data to the rest of the system, but then you'd have to trust that that is properly implemented as well.

I guess a suspicious mind could be able to check that by looking whether there is traffic between the device and the cloud servers. And the paranoid mind (who wasn't paranoid enough to not have such a device, oddly enough) could only allow the device 'net access when desiring to use its capabilities (though obviously there's no guarantee said device wouldn't have been buffering hours of conversation while offline).

What's funny is nowadays Alexa or Google Assistant will fail to wake when called but tend to trigger randomly during conversations.

Kind of like sometimes you fail to hear when someone calls your name and other times you think someone was speaking to you when they weren't?

It's not like people are perfect at doing this task either.

They do accidentally trigger very often, though. If you have Alexa devices and check the history, it often records fragments of things that were not intended for it.

If I was to have a very sensitive conversation, I'd unplug them.

But day to day it really does not concern me (or I wouldn't have four of them around the house...)

They could be under a court order to record certain keywords and we wouldn't know similiar to our telephones.

If they are not doing this now they will be.

Where is the keyword 'Alexa' recognized? On the device or back in Amazon towers? If the latter then every sounds is transmitted back to grep for 'Alexa'. Of course we believe they don't store that other noise data.

On device.

Usually done by dedicated software on a separate CPU that captures “wakeword”(mostly for technical reasons than privacy)

Both. First on the device, then in the cloud. The latter determines if it's a "false positive" or not. AFAIK there is no way to remotely start recording up to the cloud.

The cloud can push changes to the wake word (because changing it by the app is possible), it can add new ones, and the real question is rather or not amazon servers can instruct it to not turn the lights on for certain wake words.

You can see this in their home defense feature on the echos, where the echo can have the sounds of broken glass and wood breaking added as wake words so it can notify you and/or police in the event of an intruder.

I like that despite all the comments, nobody called me on "by and large the tech industry has a complete failure rate at making anything computing related secure. "

Which I figured someone would raise a fuss about. It makes me sad that we can say that and just accept it. It shouldn't be this way.

Also what’s stopping them processing advertising relevant keywords locally and then just sending a list of keywords to amazon.

Now that they've publicly claimed several times that they don't do that, I think it would really erode trust in the product if they changed the policy.

Not much beyond those here assembled.

> But I doubt they listen for anything but their keyword

> I can't imagine the NSA and other covert organizations not drooling at the chance to do just that.

Why doubt? Why imagine? You know that, for data on big corporation's servers, the NSA already gets a copy of everything - and everything is recorded forever without you really being able to delete it. Why would this be so different for voluntary self-espionage devices (Alex, Dot etc.)?

It a reasonable assumption that a lot more recording happens than is claimed.

> You know that, for data on big corporation's servers, the NSA already gets a copy of everything

This is squarely in conspiracy theory realm. Nothing in the Snowden documents or other leaks hinted at this (except for the NSA snooping on unencrypted internal links, which is useless now that everyone encrypts everything).

> Why would this be so different for voluntary self-espionage devices (Alex, Dot etc.)?

Why would a big tech company willfully infringe on their customer's privacy? How would they even do it? They have every reason to prevent it.

The Snowden documents are old now while this has become the hottest gift.

"Why would a big tech company willfully infringe on their customer's privacy?" to comply with a national security order. These companies are not lawless entities fighting for good.

I don't think NSL can even be used request the actual content of the recordings (vs. "just" the metadata) - a judge would have to sign off on that.

Now, clearly there's a lot of questionable practices like secret courts, gag orders and ineffective oversight in general, but this is still far from "for data on big corporation's servers, the NSA already gets a copy of everything", and very far from "being forced to deploy a firmware update that turns a smart home devices into a bug".

Companies act exclusively in their own interests, which I would hope includes to not engage in covert surveillance of their own customers.

This will only stay this way if we keep demanding and expecting privacy, and push for stronger oversight. But defeatist claims like "the NSA gets all data anyway" are not useful.

Stop gaslighting for the US government. You're ignoring publicly available information, universally recognized as factual.

"Why would a big tech company willfully infringe on their customer's privacy? How would they even do it? They have every reason to prevent it."

To sell advertising and recommend products?

> This is squarely in conspiracy theory realm. Nothing in the Snowden documents or other leaks hinted at this (except for the NSA snooping on unencrypted internal links, which is useless now that everyone encrypts everything).

This is absurd. Plenty of leaks clearly indicated that they succeded at breaking different types of SSL.

The working theory is that they had a practical attack on RC4, which made up a significant percentage of SSL traffic back then.

> This is squarely in conspiracy theory realm. Nothing in the Snowden documents or other leaks hinted at this

On the contrary. The leaks say this, specifically. You should read up on:

https://en.wikipedia.org/wiki/PRISM_(surveillance_program)

https://en.wikipedia.org/wiki/XKeyscore

To quote Snowden personally (on XKeyscore):

> You could read anyone's email in the world, anybody

> you've got an email address for. Any website: You can

> watch traffic to and from it. Any computer that an

> individual sits at: You can watch it. Any laptop that

> you're tracking: you can follow it as it moves from

> place to place throughout the world. It's a one-stop-

> shop for access to the NSA's information. ... You can

> tag individuals ... Let's say you work at a major

> German corporation and I want access to that network,

> I can track your username on a website on a forum

> somewhere, I can track your real name, I can track

> associations with your friends and I can build what's

> called a fingerprint, which is network activity unique

> to you, which means anywhere you go in the world,

> anywhere you try to sort of hide your online presence,

> your identity.

Reply on news.ycombinator.com