The Smart Glasses That Dox Strangers

0:03Hello, and welcome to the 404 Media Podcast, where we bring you unparalleled access to hidden worlds both online and IRR. 404 Media is a journalist-founded company and needs your support. To subscribe, go to 404media.co. As well as bonus content every single week, subscribers also get access to additional episodes where we respond to their best comments. Gain access to that content at 404media.co. I am your host, Joseph, and with me are 404 Media co-founders, Sam Cole. Hello. And Emmanuel Mayberg. Hello. All right. We have a lot to get through, and I'm mad about this story. I want to jump straight into it.

0:45This is one I wrote, and the headline is, Someone put facial recognition tech onto Meta's smart glasses to instantly dox strangers. Joe, I'm going to ask you questions so you don't have to interview yourself. I appreciate that. What is the technology exactly, and what does it do? Yes. So somebody has taken the Meta Ray-Ban glasses, which is, they are Meta's wearable and they do all sorts of, you know, AR things as Meta slash Facebook is trying to move more into that world. they do not perform facial recognition. But what somebody has done is basically tacked that capability on. And just by looking at somebody, it is able to identify it as a face, find out who that face belongs to somewhat accurately, then find their home address and their phone number and potentially information about their family members as well.

1:58And how does it do that? How does that all work? Yes. So under the hood, it is basically a hodgepodge of different technologies sort of strung together. So you'll look at somebody's face and it will then contact the facial recognition service PIMEyes. I think we're all familiar with it. We've covered it in various different contexts. For those who don't know, it's basically a facial recognition service that anybody can use and anybody can access. if you're familiar with Clearview AI, that is a very powerful facial recognition tool used primarily, I wouldn't say exclusively, primarily by law enforcement, where the police officer will point a phone at the face or they'll upload a photo, and that will grab the social media profiles that that person belongs to, because it's built on this massive data set of scraped social media images and other parts of the web as well.

3:03PimEyes is basically that, less powerful, but basically that for anybody who wants to pay 10 bucks a month or whatever it is. So these glasses, they contact that. They then use LLMs to interpret the results generated from that. Because what PimEyes does is that when you scroll through the results, It shows all these similar faces, and it also shows the URL where that face was found. So maybe you look at somebody's face, it then brings up, I don't know, their university profile on the university website or something. You click that, presumably it might have their name. These LLMs will then grab that once it's been scraped and provide sort of a summary of who this person is.

3:47Oh, they went to this school, maybe they got this award, all of that sort of thing. and then takes that name, the most important part, and it automatically sends that to a people search website, which I'm sure people have seen these. If you literally Google Find XYZ, if you're trying to identify, I don't know, a long-lost relative or something, there's all of these websites that sell access to that sort of data. And it then grabs results from that as well. And that can include the home address and the phone number. So it's not very elegant. It's not really, I suppose, highly sophisticated. It is using consumer, off-the-shelf tools and capabilities.

4:36But I think that is entirely the point here. It is entirely the point that the two people who were able to do this were two Harvard students. This was not a big tech company. This was not a surveillance firm. This was two people basically doing it in their own time. And what these two students did was they tested it on ordinary people out in the world and their friends, and they got all sorts of different reactions. And here is a voice memo that one of them sent me just about some of the reactions they got when they were actually demoing this technology out? As you can see from the video, the reactions we got were pretty insane.

5:24I was, to be honest, I actually didn't think it was insanely cool as we were building it. The main thing that made me more inspired to work on this project was that as we made incremental progress on it, and we're talking to other Harvard students, random people on the subway, that type of thing, they'd be like, dude, holy shit, this is the craziest thing I've ever seen. How do you know my mom's phone number or something like that? That's pretty weird. So yeah, I think the reactions of other people were actually a large feedback mechanism for seeing like, oh, this can be pretty impactful. So just to stress what they've done is they've built these facial recognition glasses and then they've uploaded the demo where they go and test this on unsuspecting people on the subway.

6:12There's one in particular where one of the students is looking at the results. They identify this woman as Betsy somebody. They bleep out her surname. One student then goes up to Betsy and says, oh, hey, is this you? I think we met at XYZ using information that's been scraped from her online. And you can see in the video, she is obviously receptive to that. It's obviously correct. And I also verified this because with the information in that video, I was very easily able to identify who this person was. She didn't respond to a request for comment. So I just wanted to know, like, what's your reaction to this being done to you without your consent?

6:57And I guess that sort of brings us to the motivation, which is that the students say that, well, I mean, you'll hear this in a second, but initially they did it because it's interesting and cool. They didn't bring up the dangers of this technology, which I'm sure we'll get into, and I'll ask Sam about that as well. But they brought up just how it was interesting, first of all, and only later when they were speaking to other students did they bring up sort of the context on how this technology might be used. So let's just hear some of that as well. The motivation for this was mainly because we thought it was interesting, it was cool, but as we showed people, a lot of people reacted that, oh, this is obviously really cool, like we can use this for networking.

7:53I can use this to like play pranks on my friends, make funny videos. But also a lot of people were worried like, oh, this is for stalker issues. Like if you just release this, like it'll be really bad for certain populations. Like, oh, young woman, right? Like some dude could just find some girl's home address on the train and just follow them home. So I guess like we went on to do the next step of like making a guide on how to remove yourself from these searches, like fixing your digital footprint. Sam, you've covered a lot of stalking when it comes to technology. I mean, you've done a lot of stuff on Apple AirTags, that sort of thing.

8:42What do you think of sort of that soundbite in that it initially started as a fun project and then sort of the stalking implications came later? I mean, I think it's very classic. And this is kind of what we see happen every time someone creates something and then gets a bunch of heat for it. People aren't thinking ahead about how it might impact other people other than themselves. People who have a very different threat model just moving around the world than they do. These guys probably don't ever think about being stalked or harassed or doxxed when they're just walking around outside, but it's something that especially women have to deal with all the time.

9:31Yeah, I don't know. It's very... I find it creepy and it's also just like, of course, of course this is what was going to happen with these glasses. I find the glasses is very unsettling anyway. It's like even just like the stock things that don't have any of the, you know, the facial recognition stuff in them. When someone's talking to me and they're wearing a pair, I'm like very distracted. I'm like, what are you doing? Like, are you recording this? Like what's happening? I know that there's like a light that comes on that like tells you when you're being recorded, but still it's like, I don't know.

10:05I kind of hope that these don't reach like some kind of mass saturation with consumers because I don't like interacting with them. And then this is just like another layer of like, are you running some kind of facial recognition on me right now? That's really interesting because I've never actually seen them in the wild, at least as far as I know. When it comes to you, so you've had conversations with people that presumably own them and wearing them. I mean, you don't have to get super specific, but like, did they tell you, hey, check out these glasses? Or like, did you just realize it was them?

10:42And the reason I ask is because I don't think many people even know these exist yet. And I'm just talking about the wider public that could be a victim of this who are not technology people. Yeah, I mean, so I'm referring to, because I think a new version just came out, right? And that's why it was in the news a whole lot. But there's an old version that is just like, they're just Ray-Bans. They look like regular Ray-Bans that meta-release and they have a camera in them. Yeah, and it's, I mean, they're new enough that people are like, oh my gosh look at my like check them out and it's like if you say something to somebody and you're like oh let me see your glasses they'll they're excited to to show them to you so it's not like a sneaky thing um no one's trying to like pull something over on me when they're talking to me with these but um i like it would be like having like having a conversation with someone with like they're just holding their phone like up next to their face and being like like pointing their camera at you the whole time they're talking at you even though it's not on it's not running it's like it's still strange to me um but at the same time it's like it is the technology is kind of cool like i get it i understand why people are into it um but you know stuff like this was bound to happen i guess yeah i mean i was going to bring this up later but since you you mentioned that you know you could it would be like someone just holding a phone right up like this and that's very obvious and socially we would find that very very fucking weird and that sort of leads to some of the responses I got after publishing this which is that some people said well you could do this with any camera you could do it with like a dslr you could do a phone camera you could set up a security camera and I'm like okay cool but they did it with meta ray bats and it was on their face You don't see people strapping a fucking DSLR to their face and walking because they're like, well, that guy's a freak.

12:39He's wearing a camera. But with the Ray-Bans, it is more subtle, even though there is a light there that does indicate that somebody's recording. And apparently if you cover it up, it disables the ability to do that. I don't think there is enough adoption for members of the public to realize that necessarily. Yeah. And in a place like New York, especially, it's like people are wearing and doing all kinds of weird shit all the time. So I might not necessarily on like a crowded subway train notice someone's glasses with a little light on them. But I would definitely notice if someone was pointing a camera at me, like that would be very strange and unusual.

13:17but you know I'm not like looking directly at people's glasses for a light most of the time so yeah it is it's a totally different thing I think people pretending that it's not different are I don't know like playing ignorant on purpose or something it's very strange for me to even pretend that this is the same at all as like a regular camera or a cell phone yeah i think they're reducing it to the literal tech when it's like no there's a massive social component here yeah and that's the entire point of the product right to wear it and to blend in and to not have it look like a camera on your face if that was the case we would be wearing google glass exactly and we're not wearing google glass because everybody knew it was really really strange and they looked odd uh i mean i actually know some people who hack google glass now and And I think that is cool.

14:09That's cool. Yeah. So I'm just going to play one more piece of audio. And it is about their view on how they did this responsibly. Because they did produce a guide that says how you can opt out of facial recognition services like PIMEyes, out of the people search sites. And they have not released the code necessary to do that. I mean, they've described the off-the-shelf tools that you can do and go and use, but they haven't released the code that sort of strings it all together. And there was, I imagine, and what it sounds like, a significant amount of work involved in that as well. So let me just play this last little bit of audio.

14:52What we did was the most responsible we could do. Basically, we show people that this exists, that this works, it's aware, so that everyone is aware of this now. and we're not open sourcing it so that there's some high barrier of friction. You have to know how to code. You have to spend quite a few days optimizing the system, which is what we did.

15:17I'm going to go on a tiny little bit of a rant. I'm sorry. And I'll try to keep it short. But I don't think they fully understand what they built and sort of the social contours around that and the privacy implications of it necessarily. You heard earlier on that it started as a fun, cool, interesting research project and then the implications came later. They could have, if acting responsibly, absolutely done this in a controlled environment. You know, whenever we read about, I don't know, security researchers who discover a really interesting new attack to get your password and keystrokes by listening to the rotations of the fan in your computer, ridiculous stuff that would be really, really crazy if it was out there in the real world.

16:05It's in a lab environment. They are doing it there so they don't impact anybody. They don't typically release a fully working POC proof of a concept exploit for people to be able to go do it, it's in there. If it's done on anybody, it's in that control environment with consent. Similar to when how we've bought the location of somebody's phone because it was being sold by bounty hunters. We did that with explicit consent of the target. And you're still able to, I think, very well demonstrate what is going on there. What is so unsettling about this is that you watch the video and there are real reactions from real people who are being misled.

16:52And, you know, I don't think that's responsible. And I think that is just as much as part of the story. the fact that two students can quite quickly build this and they are people who do not necessarily understand all the privacy implications the fact that they can build it is the story it's not even really about the tech it's actually about who made it and sort of how they handled that and i mean just briefly after we published one of the students said you know they said quote we tested on unknowing people to demonstrate this tech actually works. And we did it to strangers in a relatively wholesome way, acting as if we were fans of their work.

17:34I don't think you can break consent in a wholesome way. I don't think that is a responsible way to do privacy, security research. And I mean, that was reflecting some of the other coverage as well. I really can't bother to pick out particular outlets. But of course, after we broke this news, various outlets dropped on it. And some of them framed it as just a research project. They didn't bring up that there was no consent involved. They didn't bring up that it was just initially sort of an interesting, cool project. And I think that's absolutely crucial for understanding. I guess just the last, I mean, you can tell that I care about this issue a lot and it's true.

18:17And I really think it boils down to this, is that I take quite extreme steps to protect my privacy and security. As blisters will know, I don't have a mobile phone or this sort of thing. It's hard with facial recognition because you can't change your face. Well, not trivially, you know, it's hard and expensive and I don't want to, you know um and i find this sort of activity genuinely concerning and society shifting for what it might mean for um us in the future basically like if they can make it in a pretty quick way what is going to happen when somebody commercializes it streamlines it and then everybody is potentially walking around with a device that can pick out people's faces and their real identities, or even just the thought that that is possible, essentially.

19:18Anyway, rant over. Let's take a quick break, and we're going to shift gears entirely and talk about the recent hurricane and how that relates to digital archives and AI and all sorts of other sort of implications you may not immediately think of. We will be right back after this.

19:50This show is sponsored by BetterHelp. It's Halloween, which is a good time to admit that I'm still irrationally scared of the ghost stories I heard as a kid. Spooky season is a good time to face our fears, especially if they're more real than, let's say, Bloody Mary. Therapy is a great tool for facing your fears and finding ways to overcome them. Because sometimes the scariest thing is bottling up our fear and not talking it out. I've been using BetterHelp for the last month to learn strategies to deal with my anxiety and to get at the root cause of it. No matter what you're dealing with, therapy can empower you to be the best version of yourself.

20:27If you're thinking of starting therapy, give BetterHelp a try. It's online and really flexible. Just fill out a brief questionnaire to get matched with a licensed therapist and switch therapists at any time for no additional charge. Overcome your fears with BetterHelp. Visit betterhelp.com slash 404 media today to get 10 % off your first month. That's betterhelp.com slash 404 media. This episode is sponsored by Shopify. We've taken a DIY approach to everything at 404 Media, which means we didn't want to do dropship merch and wanted to handle everything ourselves. But I didn't really know how to set up an online shop.

21:10Then I found Shopify. Shopify makes it easy for me to manage our listings and inventory, sell online or in person, print shipping labels, process returns, and get paid. All with a beautiful front end for customers and a simple backend for us. It's a one-stop solution that gives us everything we need to manage our online store. Upgrade your business and get the same checkout we use with Shopify. Sign up for your$1 per month trial period at shopify.com slash media, all lowercase. Go to shopify.com slash media to upgrade your selling today. Shopify.com slash media. I love a great deal as much as the next guy.

21:53but I'm not going to crawl through a bed of hot coals just to save a few bucks. It has to be easy. No hoops, no BS. So when Mint Mobile said it was easy to get wireless for$15 a month with the purchase of a three-month plan, I called them on it. Turns out it really is that easy to get wireless for$15 a month. The longest part of the process is the time you might have to spend on hold breaking up with your old provider. Thankfully, Mint Mobile's easy-to-use website makes it super simple to buy and activate your plan using the phone you already have. To get started, go to mintmobile.com slash 404media.

22:32Right now, all three-month plans are only$15 a month, including the unlimited plan. All plans come with high-speed data and unlimited talk and text delivered on the nation's largest 5G network. You can use your own phone with any Mint Mobile plan and bring your phone number, along with all of your existing contacts. Find out how easy it is to switch to Mint Mobile and get three months of premium wireless service for$15 a month. To get this new customer offer and your new three-month premium wireless plan for just$15 a month, go to mintmobile.com slash 404media. That's mintmobile.com slash 404media.

23:11Cut your wireless bill to$15 a month at mintmobile.com slash 404media. $45 upfront payment required, equivalent to$15 a month. New customers on first three-month plan only. Speed slower above 40 gigabytes on unlimited plan. Additional taxes, fees, and restrictions apply. See Mint Mobile for details.

23:40All right, and we are back with one that Sam wrote. Hurricane Helena knocked one of the world's largest climate data archives offline um how did this first come to your attention sam um yeah first of all helene i think i was i was gonna during the break i was gonna stop and be like wait can you just check my pronunciation of this i suck so bad okay thank you for correcting to me i appreciate it yeah it's fine uh i think it's hurricane alene but um yeah so obviously like my whole entire timeline my whole feed like all my social media feeds for the last week and a half have been hurricane related and just like the the total chaos and destruction and like tragedy that's happening with um hurricane recovery in florida and carolinas and georgia and just all the kind of like, I guess like unexpected ways that this has impacted lots of different systems.

24:55I mean, we all saw like the chemical plant that's like spewing like plumes of smoke out into the air because of damage because of the hurricane. And then this story, which I saw someone who's a climate journalist tweeting about on the 29th, where the National Centers for Environmental Information, which is like a really big archive. I think it's one of the world's biggest archives of environmental data, is actually based in Asheville, North Carolina, which anyone following anything about this news lately knows that Asheville was one of the hardest hit areas by the hurricane. So it's just like it was very ironic to me almost that a huge repository and resource for environmental and climate change data was knocked out by like a once in a lifetime hurricane.

25:57and hopefully, I mean, once in a lifetime and then there's another one coming, like literally right now, but like once in a generational storm knocks out this huge data center, which, yeah, it's just, and then I went to the website to check if it was up and it was like returning a 404 error, so. Which, well, yeah, we'll get to that in a second. I mean, you just mentioned that it stores this climate data. what exactly is this data and sort of you know where does it come from and what are they measuring sort of exactly yeah so i mean it's like it's honestly it's literally everything um based on my readings about it or what i could read about it uh through like going through archives because it was all down when it wasn't down yeah well yeah it was still yes yeah well it was still down so i I mean, I learned that it contains 60 petabytes of data, which is a lot of data.

27:00It stores and releases data from everything. I think their tagline or the thing that they repeat the most is it's from the seafloor to the moon or something. It's like the depths of the ocean to space is the range of what they can collect. and they make all of it available to people. And it's not just like, oh, let's study like historical data of climate change. It's things like, like farmers use it to determine when and how to plant crops and fishermen use it to determine like where the best places are to fish like on a commercial level. So it's just, it's this thing that's like very useful on a very everyday scale for a lot of people.

27:51It's like something that the freight industry uses. And for it to be down, I think was pretty disruptive, not just for like researchers who are obviously very interested in the current hurricane season, but also how it looks historically. But people just like doing normal jobs around the world, basically, weren't able to access that information. And then this climate journalist was he had downloaded the list of billion-dollar disasters from the center before it went down because he was writing about hurricanes, I guess, and published it before it went offline. And now at this point, I don't know what the tally is at this point, but Helene is definitely probably going to be on that list.

28:45Yeah. Yeah. It's interesting that there's these like side effects basically where, I mean, maybe for in the know you would think this, but I don't think you'd necessarily assume that, oh, it's going to impact data archive and then that's going to have an impact potentially on industries and all of that sort of thing. I mean, you mentioned that to check it was down. I mean, you went there and the site and it was down, right? But like, what else did you do? I think you tried to reach out to people who work there and that sort of thing. Yeah, I mean, I checked the Wayback Machine on Internet Archive to try to figure out when it stopped scanning automatically the site because that service automatically archives lots of websites all over the internet.

29:32And that's one of them, obviously. So it stopped. I saw that it stopped doing that on the 27th. And I'm looking at it now, and it only started again on the 2nd of October. So I guess that's probably when it came back up those days. But yeah, and then I was like, well, I need someone to confirm this and or at least like tell me why maybe like why this is because maybe the building's not like flooded. Maybe there's some other reason. Maybe the power is out or something. So I was on LinkedIn just like looking for people who were who were working for and the what's the abbreviation and CEI who were working for the center who weren't based in Asheville because obviously those people don't have power either.

30:15So it's like if their workplace isn't like their workplace is underwater, they probably personally are dealing with more important things than my confirmation of their data. Yeah. Just having to go through LinkedIn and like find people who worked in like they have like a Chicago office. They have a couple of different offices. And finally, someone someone replied and passed my info on to other people and then kind of did a little telephone game from there. And then someone got back to me and was like, this is what's going on. I mean, and I guess just the last question is, did they provide an answer on what exactly happened?

30:56Like, I mean, obviously, hurricane happened. But I mean, did they get any more specific than that? Yeah, I mean, I heard some theories that I still need to go back and confirm with folks now that they actually are getting back on track. But one of them was someone who works there but wasn't directly involved in the downtown Asheville location was like, well, I think it's because the data center needs water to cool it and they don't have water. And I was like, well, that seems plausible. And then someone who works there in a more direct way, He was Dan Kowal. He's the Standards and Evaluation Branch Chief at the Chicago office.

31:44He told me that the federal building is closed and it has electricity but not water, which goes along with something someone else told me. But it's because they can't inhabit the building until they have water hooked up again. And so the whole thing was shut down until further notice. And now, I mean, it's back up now. And if you go on the site, it says NCEI in Asheville has been significantly impacted by Hurricane Helene. But it doesn't say what exactly happened. I assume the data is backed up and safe. But let me just circle back with those folks. Yeah, sure thing. uh i've listened to this name in google translate while you were talking so hurricane helena and the fuck it era of ai generated slop i know you're doing a face i'm continuing it's what google said all right i'm a gas there's no way i literally just listened to it in google anyway okay so Jason wrote this story he's not here, Emmanuel I think you edited it potentially but there and it's a fantastic piece I actually really really recommend everybody just go read the full thing, Jason's very good at articulating all of this but Emmanuel there's an AI generated image at the centre of this story why are so many people talking about it and I guess just what is that image first of all So the image is a little girl who looks very sad and is very wet.