Intelligence Scoop: Is Saudi Arabia Spying on Silicon Valley?

0:00for exclusive interviews bonus episodes ad-free listening early access to series first look at live show tickets a weekly newsletter and discounted books join the declassified club at the rest is classified.com Is Saudi Arabia spying inside Silicon Valley? And are America's tech secrets at risk? Welcome to The Rest Is Classified. I'm Gordon Carrera. And I'm David McCloskey. And this time we're looking at the question of what happens when spies meet the tech bros of Silicon Valley with a pretty wild story that's just been in the news about an engineer allegedly stealing secrets from OpenAI, the company that makes chat GPT. It's a story that involves bugging devices, toilet paper, and the question of whether Saudi Arabia

0:59might be after Silicon Valley's secrets. We'll also be telling you how to bug, or really more accurately, how not to bug your employer's conference room, explaining how Silicon Valley has been targeted over the years, including with honey traps, and how it's been targeted, yes, even by the French. But first, a word from our sponsors, HP. That's where information leaks. It's the same in business. Leaders face a constant battle against strict timelines, mismatched tools, and the worry their security isn't as tight as they think it is. Which is why HP approaches technology not as a collection of gadgets, but as an integrated system. And crucially, that protection is built in at hardware level.

2:02It runs quietly in the background, blocking threats without requiring constant specialist oversight. With a team of business advisors, HP helps businesses of all sizes find technology that fits their needs and budget. To see how HP helps businesses work securely and productively, visit hp.com forward slash classified. Trick listeners also benefit from 10% off HP business technology with the code TRICK10. Hi, this is Gary Lineker from Goalhangers. The rest is football. This episode is brought to you by Wise. It's only when you start moving money between currencies that you really think about the exchange rate, the fee, and what might be hidden away in the small print. Whether you're living abroad, paying someone overseas, or just

2:50trying to manage your money across borders, you want a fair exchange rate, an easy transfer, and no surprises along the way. WISE keeps things simple. WISE is a smart way to move the currencies you need around the globe. It works in more than 160 countries and with over 40 currencies. Most transfers arrive instantly. Wise uses the mid-market exchange rate, like the one you see on Google, with no markups or hidden fees. So when money needs to move, you can see the rate, know the fee, and get on with it. Join millions saving billions on hidden fees by downloading the WISE app today. Be smart, get WISE, T's and C's apply. than yesterday? The new Petivity Smart Feeder for Cats turns everyday meals into meaningful health insights, helping you identify changes in your cat's habits and be more proactive about

3:49their health. Plus, you can feed them from anywhere. Take the mystery out of mealtime. Visit Petivity.com, P-E-T-I-V-I-T-Y.com slash cat feeder to learn more. So David, let's get into this story of the Saudi engineer inside OpenAI and what exactly happened, which led to him getting caught. So this former engineer and Saudi national, Abdurrahman Al-Fazan, is accused of planting a sophisticated audio recording device inside a conference room at OpenAI's headquarters in San Francisco. Now, the court filing, we have a court filing and a memo from the DOJ, the Department of Justice, claiming that Al-Fazan was in frequent communication with a member of the Saudi royal family. So it kind of adds this additional twist, the prospect of Saudi state involvement in this, which we'll discuss.

4:46FBI Director Kash Patel has even weighed in. He says that Al-Fazan allegedly put this device in his quote, employer's executive conference room to secretly capture sensitive communications about the company's super intelligence technology, which I guess is now the new word we're using for AI. Yeah, I was glad to see Kash Patel, friend of the pod, be accurate in not calling it artificial intelligence, but now we're supposed to all call it super intelligence. But if we get that wrong, we apologize. So that's the accusation. But let's look at how they actually found this guy. What led them to him? September 30th, OpenAI, which has this kind of technical surveillance countermeasures team, as most big tech companies do today,

5:35are running a sweep for devices inside their headquarters. Now, they're looking, of course, for listing devices. They're looking for unwanted cameras. They're probably using a combination of radio frequency scanners, thermal imaging, kind of old school, just physically inspecting the rooms to see if they can find anything. They're also probably using something called a nonlinear junction detector, which basically floods a room with radio waves, and then it listens for the kind of the faint echo given off of a semiconductor. So this is obviously kind of complicated to do in an environment where there's a lot of different devices, but it can help point you to where every device in the room might be. And then you can find that device and figure out if it's something that should be there

6:20or that shouldn't be there. Yeah. They're going to find something. They're going to find something. The security team comes across a device that's been plugged into the back of a TV in a conference room. Now, this is a pretty important conference room. It's used for senior leadership meetings, visits to open AI by government officials. So it's a sensitive place. Yeah, it's where technical briefings happen. It's where the important stuff happens. Now, this device is apparently covered in what the press reports describe as toilet paper. Let's just pause there for a moment, David. There are many bizarre aspects to this case, but wrapping your bugging device in toilet paper, teach you that at CIA spy school?

7:01Everything that I've put in your home, Gordon, is wrapped in toilet paper. I'm going to go get a, what is it called? A non-linear junction detector right now and look for... Yeah. So let's describe what it is first and then I want to come back to the toilet paper for a second. So this is a thumb size development board that has a microprocessor that has a detachable camera, a digital microphone, and a micro SD card for storage. And its contents are encrypted, which is also quite interesting. It's been built to monitor audio, and it has a data transfer mechanism that OpenAI has said. They haven't provided more information on this, but they've said it was designed to evade detection. Now, I want it to be toilet paper that

7:45he used because it's just better. It's funny. I do wonder if he was trying to camouflage it so that it might look like it was a component on the back of the TV that had been there when it was shipped, like almost styrofoam or packaging. I kind of wonder if he was trying to make it look like that, although we should say all of the press reports have been clear that it was toilet paper. So OpenAI's security team finds the device. There's no camera in that room. So at this point, OpenAI has absolutely no idea who's planted it. Though they'll say, the security team will say they think the device has been in the room for at least two days by this point, so since the 28th of September. So the security team very ingeniously lays a trap.

8:29They remove the device, and they put a camera behind the TV so that they'll be able to see whoever comes back to check on the device, and then they wait, and someone came back. And it is Abdurrahman Alfazanan who is this saudi employee of open ai and there's a great have you seen the picture it's a great picture of him him leaning around the back of the tv isn't it and just looking at it and he looks slightly gormless i have to say as he looks at it it's literally like a chord in the act photo. You couldn't get worse than that. Yeah. He kind of has his mouth is slightly agape and he's kind of, he just kind of like, oh, it's like that. He's just like, you can see. It's like the moment you realise.

9:14Where's my toilet paper? Yeah, exactly. Where's my toilet paper? Oh, there's a camera. Uh-oh. So he knows he's in trouble. So who is he? Let's work out who he is. So he's 35. He's an engineer who is he? Let's work out who he is. So he's 35. He's an engineer who joined OpenAI back in July of 2025. Came to the US in 2009. He studied at MIT, at UC Berkeley, two of the very top universities in the US for tech subjects. Spent about a decade in Silicon Valley tech. He's a Saudi national, US permanent resident. His LinkedIn profile lists his role as engineering lead on OpenAI's AI agents platform. And that profile says that he's lived out in California. So out in the Bay Area working in tech since 2015. So pretty experienced. He's got a great resume. He's also

10:06a partner at a Saudi venture capital firm. He'd previously had some experience at Meta. So, you know. Yeah, he's qualified. He's a very qualified tech bro, is what he is. And Al-Fazan is the gentleman caught inspecting the back of that television in that wonderful photo that has been released by the DOJ. 20 minutes after he stares mouth agape at his missing toilet paper, Alphazon goes back to his desk. And I should say, I'm piecing some of this together from the press reports because we don't have the sort of beat by beat chronology, but this is, I think, very close to what happened. So Alphazon goes back after seeing the dev board is gone. And then he reports, he's like, my bag's missing. Now, I think as he's in the room, the OpenAI security team has decided to intrude on his bag.

11:04Lift his bag. Theyude on his bag. Lift his bag. They've lifted his bag. That bag has a second recording device, which the security team confiscates. And OpenAI has also said that they have evidence he intended to deploy more devices throughout its offices. Now, security seems to have put his bag then back in the lost and found. And AlphaZon goes goes picks it up and he complains that he's missing a credit card and cash but he doesn't say anything about the device which the security team has confiscated from that bag because you'd be a pretty inept spy to go oh i i've lost my bag oh and in it was a secret bugging device I was using to target the company. So at least he passes that one test of not kind of saying that to the lost and found team.

11:52But he's in trouble. What I feel like he should have done is given that the missing bug and the bag suddenly disappearing and going into the lost and found, and now you're missing the second recording device, given that that happens in the span of no more than a couple hours, if I were him, I would have gotten on a flight to Riyadh. Yeah. I was going to say, that's the moment you know the game is up. Right. I mean, just what are you doing? I mean, at this point, you should know, you shouldn't be complaining about a missing credit card. They're onto you, right? Now, it does seem that in addition to the camera trap and taking his bag, the OpenAI security team did at least three other things to figure out that it was Alphazan.

12:41So first, very simply, there's camera footage not in the room, but in the hallways in the corridor outside the room. And they go back and they look at all of the different people who have passed through those corridors. And those videos show that Al-Fazan was near the room several times, although he's not authorized to go into the room or to attend meetings inside it, right? So he's kind of like, why are you there? Second, they go through the internal productivity agent logs they have. So they, you know, open AI engineers will have their own kind of in-house productivity agent they're working with. And the security team goes through the logs to see, has anyone been querying their productivity

13:23agents about this particular room or the meetings that are being held in this room? And Alpha's on hat. He's virtually typing into his agent bot. When is there a sensitive meeting in this room? I mean, that seems to be basically what he's looking up. Not that smart. On the tech firm's own tech inside its network. Has he not heard of surveillance capitalism? Does he not understand how this stuff works? Yeah. Okay. The third thing that the security team does, which I think this is the best part, after they find the device, they obviously analyze it. So they've got the actual brand name of this, you know, microprocessor. They've got the device's specs. They realize it's been modified. We'll talk a little bit more about that in a second.

14:11And so what they do is they take all of the specs and modifications and they put those into an AI model that then goes and searches ChatGPT to see if anyone's chats had been geared toward constructing a device like the one that they found. And we should say actually, and I'm still trying to figure this out is, was this done on sort of an internal chat GPT? It seems to have been done on his wife's chat GPT account, which we'll talk about. And so there is this question I think of, is OpenAI then able to search and find, like, are they searching from the backend and figuring out what you're looking at and sort of profile? I mean, it's a very interesting kind of question. How much of your chat activity gets

14:56saved? That's the question raised by it. But they look and they find that his wife has been looking, or at least we should be careful about this. His wife's account. Yes. So his wife's chat GPT account had chats and prompts about building an encrypted audio recording device. And it included researching things like, how do you build a continuous audio recorder that saves compressed WAV files to an SD card? That's one. How do I add speech-triggered recording and live listening? How do I add browser-based remote controls? How do I encrypt the card and later decry in addition to all the device specifications, you know, fairly damning evidence that this is the gentleman who constructed said device.

15:51So he's arrested. Al-Fazan is arrested by the FBI on the 9th of October. Interestingly, the arrest comes three days before Al-Fazan was due to fly back to Riyadh with his wife and son on a one-way itinerary. Which would suggest that they were watching him trying to work out more and they might have wanted to wait more. But when you see, I mean, the FBI often do this. They look at someone's travel itinerary. If they see they're about to travel, especially on a one-way ticket, you've got to pick them up before they go. So it looks like that's what happened here. It is interesting because Al-Fazan, in this sort of period after the arrest, he's not speaking to law enforcement. He's like, give me my lawyer, which is sensible.

16:37But his wife does speak to law enforcement. And she says, apparently, I don't know how, I mean, this is just, it seems wild to me. She says, oh yeah, I have seen him using my chat GPT account on his laptop, which is again, really quite damning. She says that, problematic. She says that she and her son are moving back to Saudi Arabia permanently and that Al-Fazan had planned to return after a couple months to keep working at OpenAI. It does seem like this coincided with his paternity leave. His son's very young, I think three months old or something like that. So his wife is saying, yeah, we're going to Saudi, he's going to come back and keep working at OpenAI, but he does eventually intend to move to Saudi Arabia long term.

17:25And then it's interesting because just after the arrest, his wife and son do go to Saudi Arabia and they're rebooked on a one-way flight to Jeddah for October the 11th. And the DOJ, the Department of Justice memo, says that this ticket was booked and paid for by the Saudi government. Curious. It is curious. That flight is canceled for some reason. And then they rebooked the wife and son on a ticket, a one-way business class ticket to Doha on the 10th of October that is paid for. Now, I think this has been a little bit deceptively reported in the press because the press, and I believe the DOJ memo as well, say that this ticket was paid for by a, quote, member of the Saudi military. But it seems to have been her brother-in-law who's an official in the Saudi Navy, I believe.

18:21The brother-in-law is like, oh my gosh, my brother-in-law has been arrested and I need to bring my sister-in-law back home. But either way, she and their kid are out within 36 hours of the arrest. They're gone. That's right. And the FBI is continuing to gather more evidence from Alphazan's home, from his iPhone. They find more recording devices at his home. On his iPhone, investigators find photos of at least one other concealed recording device, the input output panels on TVs and other equipment around open AI buildings, including another conference room where sensitive meetings happen. So it's, again, probably kind of reconnaissance for planting more devices. And then they have this, you know, in a deleted September note that he had made to

19:09himself on the phone, they find links to other recording devices, including ones made by a company called Seed Studio, which will play a role in the story when we look a bit more in depth at the actual recording device. There's also a great folder titled Riyadh Tasks, in which he is trying to find an apartment for his wife and son, a job for his wife, getting them IDs and passports. There's an open AI leave note, in which he suggests a, quote, early pickup of the devices and extending his parental leave to full. And the prosecution, the DOJ, has read this as a plan to retrieve the recordings before he leaves for Saudi. It's not great in terms of... It's not a great look. He's not covered his tracks very well. Not clear from the DOJ memos about what conversations were captured or might have been or what recordings

20:12were shared or what happened to them. You know, OpenAI says it's got no indication he improperly accessed or took model weight source code or customer data. That's the kind of sensitive commercial charge. I mean, also no formal espionage charge from the government that's interesting uh there have been other times and we might come to those where there are espionage charges for cases like this so all of that suggests this is a guy trying to you know get hold of material but it does leave the motive open as to the extent to which he really was acting on his own or acting as a spy. So I guess the question is, what are the signs that this was some kind of Saudi intelligence operation? So we have the purchase of the flights for the

21:04wife and son by the Saudi government and then flights canceled. But then maybe it's her brother-in-law, a member of the Saudi Navy. But this is not the only connection. Crucially, there are WhatsApp messages between Alfazon and a member of the Saudi royal family, unspecified, unnamed in the DOJ memo, which is frustrating. And these messages run from 2025 up to 2026, most recently on the 5th of October, right? So just a few days before he's arrested. The unnamed royal has attempted to recruit Al-Fazan for a Saudi AI initiative. It seems like the flavor of a lot of these messages is the royal has attempted to recruit Al-Fazan for a Saudi AI initiative. It seems like the flavor of a lot of these messages is the royal is trying to convince Al-Fazan to come back and leave OpenAI and work in Saudi on Saudi AI initiatives.

21:57I mean, in May 2026, for example, the royal writes, Saudi needs you more than the US. Al-Fazan replies with the hundred percent red thingy you know in his in his text right so this is the kind of kind of the theme of the banter back and forth between the two of them and it sounds like they alfazon is is saying he wants to do it but it looks like i mean he's waiting for some stock that he's got equity to vest. Which, you know, happens in Silicon Valley and elsewhere. Which makes sense. Yeah, you get trapped. Because I think it's like $2.2 million that he's got, which is about to vest if he waits long enough. I mean, his regular salary was $385,000, which is pretty good. So it looks like, you know, the only thing that's holding him back

22:41is him waiting to get that money out of OpenAI. But he's making his preparations to go. And he's also got another folder, hasn't he, called Saudi GovDocs, which he accesses just before this all heats up. Yeah, the Saudi GovDocs folder is not a good look when you're picked up by the FBI on suspicion of having planted a recording device inside OpenAI. Now, Alfazan, he also has asked his AI productivity agent for information about Saudi engagements with OpenAI, including who from the firm would attend high-level informal meetings with Saudi ministers. He's searching for a list of planned open AI visits to Saudi and the like. So we're not making any allegations here on the rest is classified, right? But when you add these things

23:33up, you could make the argument that this looks like a guy who's helping people who work inside the Saudi government and who work inside Saudi Arabia's AI sector learn what's going on inside OpenAI and feed that back to them? Yeah, I guess there are two ways of looking at this, though. I mean, one is he has been tasked to do that, you know, by the Saudi government or by this prince or by whoever to say, go to OpenAI, get stuff, bring it back to help us. And that would be if you like espionage or something like espionage. But it's interesting, there isn't an espionage charge in this case. It is interesting there's not a charge. Yeah. The other option is, this is a guy who everyone knows has access

24:22to very, very sensitive, useful information. And they are saying, come work for us, knowing he may bring it. And he's thinking, what can I take and what can I get hold of that will give me value stroke wealth, riches when I go back? So I think that's, in my mind, the kind of question which is left slightly open about this. I don't know. What's your sense? I mean, OpenAI has been very clear that this was not an attempt to steal... It wasn't really an attempt to steal IP, like technical IP from the organization. The really sensitive stuff, yeah. The really sensitive stuff. He's tried to record audio in rooms around OpenAI. That makes me think that what, and given his interest in the Saudi meetings at OpenAI

25:17and probably the sort of, you know, as in any organization, there's going to be prep meetings in-house before you have those meetings. And I wonder if part of what he's after is to get a sense of the room as the kind of principals from the OpenAI side are preparing to meet with their Saudi counterparts. He can get the kind of inside scoop about how they're thinking about the meeting, what they want out of it, the kind of unvarnished version in-house prior to actually meeting the Saudis that Al-Fazan could pass. And I don't think that that would have to be tasked. I think, honestly, it makes a lot of sense to me that if he's got paternity leave, they're going to spend it in Saudi. He wants to be there eventually. He might've thought, I want to run this bugging operation,

26:09get some useful stuff that when I go back, I can share it with people, use that to increase my standing and eventually whatever role I could get once my stock vests and I can leave San Francisco. So I don't think there needs to be, I think I'm with you, I don't think there needs to be a tasking here. I think he could, it makes sense given the tech he's got and given the relationships he's got that he's just doing this for his own personal benefit. There is though this history of this kind of targeting of Silicon Valley specifically by foreign states. And of course, the reason is that the US, Silicon Valley is at the leading edge of global technology. Other countries know they're behind, therefore,

26:50they want the secrets. And in the Cold War, the KGB had this thing called Line X, whose whole job was to steal science and technology. But then it's also interesting, if you go back through the history books, you do come up with other cases where it's also countries a bit like perhaps the saudis where they're kind of friends kind of you know like in 1982 there's this joint fbi and ibm team from ibm the computing company who set up a fake consulting company which is designed to ensnare soviet spies who are looking for high-tech equipment. And it ends up catching a group of executives from the Japanese firm Hitachi, who was stuffing confidential information about disk drives. I don't know if you remember those. You might be too young to remember those.

27:34No, I remember. Gordon. Gordon. Gordon. I remember disk drives. Okay. Very good. Three and a half inch. Anyway. But then also the French. I mean, we can't resist the reference to the French. The French. we can't resist the reference. They love commercial espionage. They love them some commercial espionage. So they get caught planting agents inside IBM and also Texas Instruments, which was a big computing firm back in the day, sending details back to their French companies. And there's this one story like May 1991, a guard at the home of a Texas Instruments executive sees two men throw trash bags into the back of their van. And the guard writes down the van's license plate number and it gets traced to the French consulate, I think in that case in Houston.

28:18And the consulate are like, we were just collecting grass clippings. I mean, why the French consulate needs to collect grass clippings, you know, but they were clearly doing trash collection, trying to see if there are any documents there. So I think, you know, you definitely get a sense that states, including sometimes friendly ones, are after Silicon Valley's secrets. I mean, we did a couple of episodes as well, didn't we, on Operation Aurora, which was back in 2009 when Chinese spies hacked into Google, which was trying to steal their source code. So Silicon Valley is a target, isn't it? But even with that world of cyber espionage, I still think it's this issue of insiders in recent years that's particularly interesting. And I think AI and companies like OpenAI make it

29:08particularly important as a target because I think of the national security implications. I think it's really put the focus on Silicon Valley now. Well, maybe there, let's take a break. When we come back, we'll dive deep into the fears of bugging, of insider threat in top tech companies today. And we'll explain just how easy it is to bug your employer. We'll see you after the break. Shaq here. What's it like to buy a car from Carvana 100% online and get it delivered to your door? Let me tell you. Your car is here, Mr. Shaq. Thank you, Mr. Carvana guy. Can't wait to take it for a spin. Only the best for you, Mr. Shaq. Hey, have you always been so Let me tell you. So buy your car today on Carvana. Delivery fees may apply. See our seven-day return policy at Carvana.com.

30:11Well, welcome back. We are looking at spies in Silicon Valley, and we're going to get into the details of bugging devices. But first, there are some signs that big AI labs are really starting to get worried about security. Yeah, that's right. I was just been reading Karen Howe's very good book, Empire of AI, which is about open AI. So the company targeted here. And it is interesting that from the very start, they are worried a bit about bugging devices, but it's primarily from other rival labs and tech moguls rather than foreign states. I mean, it's so interesting because Sam Altman, the boss of OpenAI, compares slightly grandly what they're doing at OpenAI to the Manhattan Project, you know, building the nuclear bomb.

30:58But they don't really act like that in the early days, you know, as if there's a risk of foreign spies stealing stuff. Are they as equally penetrated as the Manhattan Project was, Gordon, do you think? The Klaus Fuchs, back to our early episodes on Klaus Fuchs. But at least the Manhattan Project, you know, was pretty secret, had quite a lot of security. You know, here, pretty secret. Well, the Soviets knew about it. They did. That's what I mean. It's pretty secret. Pretty secret. Well, the Soviets knew about it. They did. They did. That's what I mean. It's pretty secret. But it's interesting because OpenAI, Sam Altman has a breakdown of relationships, and there's been a recent court case about this, with the original founder, Elon Musk.

31:37And so at one point, Sam Altman secretly commissions an electronic counter-surveillance audit. So a bit like what caught this Saudi. But he's looking for bugs that Elon Musk might have planted to spy on open AI. So the kind of paranoia is there with other tech companies. And I mean, you see it, one of the other top scientists who's now also left, Ilya Sutskeva, he mused to colleagues about what he should do if his hand got cut off to be used in a palm scanner for unlocking open AI secrets. I mean, that's a man who's watched too many spy thrillers, I think, isn't it? So he was wondering what he would do after his hand got cut off? I think go to the hospital is probably the answer and call the police.

32:25And then they have these, you know, they have keypads on doors, which are programmed to have distress passwords, which are special codes that you could punch in to trigger a secret alarm, alert the security personnel if you were being forced to open the door, in other words, by someone. So, you know, they start to worry a bit about security but in a fairly amateurish way and i think it is interesting because you know going back to this saudi case the problem is these are open growing firms they're also very international i mean there's stuff full of people from around the world because they're competing for the best talent from around the world and so the real risk is actually of insiders. It's the insider threat.

33:08It's someone coming into the firm, you know, and taking the secrets, whether for a foreign state or whether for a foreign company. And there have been lots of lawsuits in Silicon Valley about people, you know, leaving one tech firm, going to another, and then, you know, effectively taking, you know, commercial secrets and IP with them. So this story does feel like it's partly perhaps within that world. But I guess the other thing that's interesting about it is the bugging part, because that's what an insider allows you to do. Rather than having to do cyber espionage or anything like that, you can put a bug inside an office and potentially gain some interesting information, can't you? And I mean, that seems to have become, if you read this

33:53account of this OpenAI case, a bit easier to develop what would have been spy tech. Yeah, no, absolutely. I think this case, despite the very sort of amateur Inspector Clouseau aspects, you know, from the toilet paper to the missing bag and just leaving this event's digital trail of evidence, you know, all along the way, I mean, it really does raise this big question, which is, how easy is it to do this? I mean, a lot of the stories that we do on the pod could be Cold War stories about spy agencies using listing devices and bugs. But I think there's a sense that we're now, we're living in a world where a lot of this stuff is commercially available today. And I think it raises this question of, well, just how easy would it be to do this, right?

34:46And I think maybe the place to start here with the sort of the Alfazon case is let's just take a quick look at what he was actually using to put, to bug this conference room or attempt to bug this conference room. The court documents that are filed by prosecutors identified the recording device as a Seed XIAO ESP32-S3 Sense, Gordon. Do you know what that is? Yeah, I've got a bag full of them. Gordon's got a sack full of those things behind him on screen. But it's cheap. It's cheap, isn't it? It's cheap. And here's one key point, which we should just emphasize. He didn't put malware on OpenAI's network, right? He's trying to bug a room. So we're talking about, in this section, we were talking about listening devices.

35:30We're not talking about how you would actually get onto a computer network and take IP secrets, mess with data, whatever it might be. This device, this microprocessor that he has that's rigged up with a microphone and a camera. It sells for $13.90 in Seed's own store, and it sells for about $23.50 on Amazon. It's a pretty low-end camera, roughly the resolution of the first iPhone. Doesn't seem like Al Fasan was particularly interested in the video. He wanted it for the mic. And the device itself, as those prices probably suggest, is not particularly sophisticated. What is slightly more sophisticated is the custom firmware modifications that Alphazot made to the device, and also the fact that he made it

36:17encrypted. So he encrypts the stored audio on the card, so all the SD cards, only he can access it. As we had discussed before the break, he's using his wife's ChatGPT account to actually help him make the device and configure it. Using ChatGPT to try and hack OpenAI, which makes ChatGPT. Nice move. But it teaches him. He's not having it. It clearly works. It's not that many guardrails on it about how to kind of make it effective. I mean, it's so interesting, isn't it? I mean, that is the point to some extent about AI and about these large language models, is they can make it easier to do things which were technically harder before. It's that ability to ask it questions and teach you, well, how do I make it speech triggered?

37:04How do I improve the speech triggered? How do I improve the audio quality? How do I control it remotely? He's using ChatGPT to teach him how to take a kind of regular camera that you could buy, as you said, from Amazon for not very much money and turn it into a slightly more sophisticated piece of spy kit. Yeah, I think it does. It raises this question of, well, okay, does this then show us that bugging, like so many other aspects of the spy business, has kind of jumped the tracks, right? Is it now commercially available, publicly available, watcher, can you run your very own sort of super secret squirrel bugging operation? We would not endorse such a move. We would not endorse it. We're not encouraging people to break the law and spy on their own companies.

37:55But if they were to do so. We're going to explain how to do it. Yeah. So, I mean, again, there's this sense, I think, that this is a new phenomenon, commercially available bugging equipment, and it's not. I mean, Newsweek magazine, I was looking as we were preparing for this episode, ran a story in 1975 about what they called the debugging mania that was then sweeping corporate America, how private counter surveillance firms were charging thousands of dollars to conduct sweeps for bugs. Newsday ran a story in 1990 called Spies Like Us about what was then a hundred million dollar a year consumer surveillance industry, the use of teddy bear cams to watch homes and make sure the babysitter's behaving. So the consumer spy tech

38:43of the sort purchased by AlphaZone for this operation is not really a new thing. Yeah, I mean, my favorite bugging story is someone telling me that they discovered a bug in what's called the senior common room of a well-known Cambridge college. Because there was a massive row going on between the professors and the head of that college. And the bug had been placed exactly at the spot where all the senior professors went to have their surreptitious conversations. And this was something which was being used a few years ago, but in a Cambridge College. So you're right, This is something kind of quite widely used and available. That's right. And I think it's helpful to break down some of the components of this type of

39:31operation because these are going to be consistent whatever the time period or technology. And I think it can help frame a bit of what's new and what's not and what pieces of these sorts of operations still remain very challenging, right? Despite the proliferation of all this consumer tech. So there's basically four components, right? There's access. You need access. You need access to the room or to the phone, whatever it might be, to plant the device. You need the hardware. That's the physical device, including how is it powered? Is it battery operated or is it actually wired into something? That's two. Three, concealment and storage. How do you miniaturize it or camouflage it so it can't be detected? And then this is important, how do you actually get the data out and keep the device

40:16concealed? There's some different trade-offs there. And then analysis. This is a part that we don't think about a lot when we think about these operations, but how do you actually make sense of what you have and use it to actually gather intelligence or secrets that are valuable? So if we take that model and then apply it to the open AI case, you get a pretty interesting example of the way it has been democratized effectively, the ability to do this kind of bugging operation. Yeah, I think it does. I think it's way more democratized. It's cheaper. It's easier. And yet, I think we'll come back to this once we run through this kind of lens for the OpenAI case. I think there are still really significant challenges that private individuals or even

41:02firms have and that are really difficult to overcome to run a really effective bugging operation. So we'll come back to that. But I mean, so start with access, right? In the OpenAI case, Alphazon, you know, he doesn't have authority to be in that room, but he's an OpenAI employee, which means he can go around the headquarters, he has access to a lot of the network. You network. Physical access is still unique. It's still very important. That has not really been democratized. No. I mean, because the interesting example is if we did a while back Operation Overt, which was the airline plot being followed by MI5. And it was so interesting there because you had MI5 having to effectively break in legally to someone's flat to be able to plant the bugging devices in that case. So if you don't have the access, that is what you have to do. And that is obviously dangerous and potentially, if concealment, right? Because in the Operation Overt case, because they could get persistent access for hours and hours, they could install the microphones and cameras in places where they were connected to the home's power source.

42:20They were completely hidden away. Whereas, you know, in Elphazan's case, because he has sort of has access to the room, but really shouldn't be there. He's going in and out and just kind of sticking something in the back of a TV, right? Because he can't hide it more effectively, right? Now, I mean, there are commercially available, you know, microphones and some other kind of creative devices that can give you access to sound inside a room at longer distances. But access is still a real problem. And I think much of the tech we're talking about really hasn't made that easier to solve. Now, that's access. So when you go to hardware, I think this is where it has become really seriously democratized. Because basically,

43:03every iPhone is a high quality recording device. You can buy high quality microphones for basically peanuts online. They decline in cost by around 10% every year. So just keep getting cheaper and cheaper. Concealment and storage, I think this is the most interesting part because in OpenAI case, we have toilet paper used as camouflage for a dev board that's plugged into a TV, which isn't really great. I think we would both agree that this is not the ideal setup. Gold standard. Yeah. Alphazon's mistakes, I would say it's probably true that concealment is easier today. Because I think a few decades ago, we're talking about a commercially available bug that you would get if you were conducting industrial espionage. It's probably a small battery-operated radio transmitter,

43:57and it's probably broadcasting almost constantly. So it's a little bit larger, and it's easier to detect because it's constantly exfiltrating that data. And you have a limited lifespan with the battery, so it's easier to detect physically and then it's easier to detect with radio frequency scanner. But today, I mean, AlphaZone's essentially thumb-sized board has a mic, a processor, a wireless radio, dozens of gigabytes of storage. It runs off a USB port. So the devices themselves are much smaller and more compact and easier to hide. Yeah. So easier to hide, especially amongst all the kind of electronics and things that you have in daily life. There's more stuff around, isn't there? They're more powerful as well, aren't they?

44:44I mean, you can store more and do more with these devices than the bugs of the past. I think that's crucial, isn't it? That is huge because there is this trade-off between the device transmitting and then being able to store the data. And then you could go back back and actually like in Alphazan's case, even if the device had not been capable of transmitting, it can store 64 gigs of data, I believe. And so I mean, that's that's weeks of audio files, right? So he could he could go back like three weeks later, and switch out the SD card. And, you know, the thing isn't transmitting, which makes it much harder to detect. And that storage is really a critical piece of this because hard drive prices, especially when you compare this

45:31to the late Cold War, have absolutely collapsed. We've gone from it requiring a half million dollars to store a gigabyte in the 80s down to like three cents by the late 2010s, right? So- Storage is cheap. Storage is very, very cheap. And if AlphaZone had been doing this in the Cold War, it would have taken him millions of dollars and the cubic space equivalent to like 64 refrigerators to store this information. And he's got it all on this thumb-sized deadboard. So the storage is a passive part of the story because it's just cost-effective to do it. And then the last point on the analysis, I think this is really important, and it's one we often neglect. If you thought about a surveillance operation, again, let's just use the Cold War

46:16as an example. Someone has to listen to all this stuff. If you're bugging a room, someone has to make sense of all of the different meetings that are happening at this room. What's important? What's not? You've got to create transcripts that other people can read. So that kind of audio surveillance was very, very labor intensive to find the secrets out. No, I mean, I remember talking to the FBI team who investigated the ghost stories, the Russian illegals who got picked up in 2010. And they still at that point had to listen. They had teams just listening to hours of tapes to try and find the relevant and interesting conversations. But I guess what we know AI now is very good at is things like transcription.

47:03And also you can see he's also looking for things like voice triggering so that it only starts the recording when there is something to listen to. So you can see there, can't you, where the technology, and it's kind of technology we all use. I mean, we use transcription software all the time, is really helping with something like this. If you're doing a private bugging operation, you don't have that many resources. Yeah. I mean, imagine that Alphazad had gotten a couple of weeks of audio off of that bug and he's in a pre-AI era and he's trying to make sense of it. Number one, he'd have to go out and actually probably hire someone to do the transcription. And that would be very risky because they might wonder what in

47:46the world is this, right? So you've got to sit there all night listening to this stuff. Just listening to this thing, right? I mean, in September, Meta launched an open source transcription model that charges about 18 cents per hour of audio you put in. So AlphaZone could do this for basically nothing. It's almost instantaneous. And then he could use IDLLM he wants to kind of mine the useful information out of those transcripts, right? So bugging operations, cheaper, easier than ever before, although the access point I think is still hard. But I actually think what's harder now is not getting an operation going, it's keeping one clandestine over the long run. That's the hard bit, not getting caught.

48:34It's easier to run a bugging operation, but it might also be easier to find one. Well, so his case, again, I think it makes this point very well because we have small, cheap cameras everywhere. I mean, Alphazon had a small, cheap camera on the microprocessor he plugged into the television. But he was also caught by small, cheap cameras that OpenAI had installed behind the television, but then even more importantly, had installed around its offices to keep tabs on who was where. His design research sat in his own chat GPT account, right? Back to surveillance capitalism, catches you in the end, your digital trail will get you. That's right. I mean, so we actually have a case here where the sort of AI surveillance world has

49:20come full circle because Alphazad uses AI to help him customize the firmware for his bug. And then OpenAI uses its AI models to analyze the bug and to then catch Alphazad. So it's like this self-licking ice cream cone of AI, you know, creating the bug and then finding the bug afterward. And then of course, you know, he's got on his phone, he's got all this stuff. He's got photos on his phone, which they can recover, and he's even got a deleted note on his phone and other notes, which, again, even if he's tried to delete it, there's still a digital trail there, you know, as well as purchase records, all this kind of other stuff. The digital exhaust we leave behind, I guess, makes it harder if someone is onto you

50:06to hide the fact that you're running this kind of operation. Yeah. And as we discussed at the beginning, I mean, OpenAI has a security team that's running counter surveillance and sweeps around its headquarters on a regular basis. This has moved from the realm of just public sector intelligence agencies, defense contractors, into something that is done at large corporations that have sensitive IP, like most tech firms do. So this has become very common. This was not common 30 or 40 years ago at most companies. It's become much more common today. The last point I would make on this is that if the tech landscape has sort of equalized between the public sector, private sector, in terms of the know-how and in terms of

50:54the hardware that's available, I do think it's still true that the state, I think the story makes this point clear. A state has scale and legal advantages over private individuals and firms. And those advantages make it easier for a state to hide its tracks when it conducts these operations. It's much harder for you to hide your tracks to conceal what you're doing. A state has, you know, they can build listing devices into buildings. They can run supply chain operations against big hardware shipments. States run institutions like the FBI, the CIA, the NSA that don't keep their research on bugging technology in their private chat GPT account, right? So there's a clandestinity afforded by secret institutions. And we talked about the Operation Overt case. States can,

51:46in some cases, legally break in and create the access necessary for long-term concealment in a way that private individuals simply cannot. So, I mean, in other words, it's got easier to do your own private bugging operation of an employer, but it's still not as easy as it is for MI5 or the FBI. I mean, it's interesting, though. There's been this sudden rush of worries in Silicon Valley about spies after this story. But actually, I mean, if you read some of the press reports around there, it seems they're more worried about honey traps than bugging, which I think tells you something about people's interests. I was reading a piece in the San Francisco Standard. It's narcissism. It's Silicon Valley narcissism.

52:23The piece starts, It's already hard enough to make friends, find a romantic partner and network in San Francisco. Now the city's tech workers have another factor to consider. Is this person being nice to me because I'm cool and interesting or are they a spy? Foreign intelligence agents are most often young, attractive women, experts told the standard. And it does seem like actually OpenAI has been running presentations to warn people about people approaching them, including HoneyTrap. So I love the fact, you know, this was like a Saudi engineer running a bugging operation. But what everyone wants to think about and worry about is the HoneyTrap. Is that true? I mean, should they actually be concerned?

53:02I mean, I guess in theory, yes. But I mean, what do you make of that? male and nerdy crowd i'm sorry and i say that is in some ways as a nerd myself but you know who who and therefore i think you could you can see the risk of honey traps there but actually you know i haven't seen that many examples of it in silicon valley and i've seen more examples of tech guys you know sticking in usbs or stealing the data themselves rather than being honey trapped so it comes back to the kind of you know what's the myth and the fantasy of spying and what it's like and what's the reality? Well, I think maybe that's a good place to leave it, Gordon. I think regardless of whether we're talking about honey traps or we're talking about bugging operations,

54:00even poorly run bugging operations like the one Al-Fazan ran. Obviously, there is a significant insider threat concern that's real in Silicon Valley. And there are plenty of actors, be they for private gain or foreign intelligence agencies, that are after really valuable secrets, IP, and sort of the plans and intentions of these tech bros. The defense, I think, I mean, that these firms will need to take is not, again, it's not just against the Russians or the Chinese, but to some degree, it's against people who are working inside their very own buildings, like Abdulrahman Al-Fazan. And I guess as we finish, that's a perfect way to remind people about perhaps one of the great insider threats,

54:47Kim Philby, our last episode on the man who got into MI6 and spilled its secrets coming out next week. Also with the bonus episode alongside it for club members about Philby himself talking about his career from this amazing tape I uncovered from the Stasi archives, plenty of other great bonus material that's been there with that Philby series as well to listen to, and the next series coming up after that, The CIA and Watergate. It's a story which has actually been in the news recently. It's been referenced by none other than J.D. Vance and some other people talking about the deep state and that awful conspiracy against Richard Nixon. It's a bugging story, too. And it's a bugging story, too.

55:33So everything comes together at the end. So do stay tuned for that.