Did OpenAI Just Catch a Spy?

[MUSIC] Big AI companies have a lot to worry about these days. There's a chip shortage, there's agents going rogue. There's even concerns over their tech destroying humanity. And on top of all that, there's also a grooming threat of spies. [MUSIC] The way we've heard security experts describe this kind of stuff is, it's like a cat and mouse game, which is, you know, basically, constant. You know, it's never ending. Our colleague Maxwell Zepp has been reporting on a recent case at OpenAI, where a bug was found planted in a conference room. These rooms are used to discuss top secret research and meet with government officials. And on the last day of September, OpenAI conducted a sweep. They would just kind of take a look around after meetings, after people had been in there, see if people had left anything, kind of look under the tables, under the chairs, behind, you know, the corners. And that's actually a key part of this story of how they found the device in question. >> The device in question was a microphone. >> A small microphone attached to a micro SD card. And OpenAI found it plugged into the back of a wall mounted TV. And actually it was wrapped in toilet paper. And almost immediately realized that, you know, this is a recording device that we, the company had not put here, that someone else had put. And it kind of, you know, immediately set off alarm bells. >> Within a week, OpenAI said they found the culprit, and the FBI made an arrest. So OpenAI and the Department of Justice are certainly treating this like a big deal. It not only raised security questions about what's going on inside of OpenAI, but also just the broader industry of how they're protecting the secrets behind the AI battles that they're working on, and also who is trying to get them. >> Welcome to the journal, our show about money, business, and power. >> I'm Emani Moiz. It's Thursday, October 22nd. >> Coming up on the show, was there a spy at OpenAI? >> The man being accused of bugging OpenAI was an employee of the company. His name is Abdul Rahman Al-Fazan. >> So were there any red flags before this incident in his background? >> So not really, he had a pretty clean record before this. And we've learned a lot about Al-Fazan through way of court documents that have been released in his case in the last week. Al-Fazan's a 35-year-old Saudi national who came to America in 2011. Since then, he's gotten married, had a kid, and collected a star-studded resume working in Silicon Valley. >> He studied at MIT, and right out of college, he got a job at Metta working as an engineer. And then he later worked at a startup that was backed by Y Combinator and Andreessen Horowitz. And most recently, he joined OpenAI. And something we've learned about Al-Fazan through way of these court documents is that he was a very popular guy. There were a lot of people after his charge was made public. A lot of people wrote into the court to speak on behalf of Al-Fazan. He got more than 50 letters from friends of his. >> Wow. >> Eight of which we know their names. Some of these people work his co-workers at Metta. Some of them people who work at Tesla. There's one person who works at Apple in here. And they spoke incredibly highly of him. People describe him as hardworking. Someone described him as a role model. >> One person said that he was the person that Saudis would turn to when they first arrived in the Bay Area. >> So offering advice to young engineers, people looking to have a similarly incredible career in tech as him. >> So yeah, I mean, this was someone with clear and deep roots to Saudi Arabia, but also really acting as kind of an organizer of this community here in the Bay and a resource for a lot of people. >> Al-Fazan's attorney said the allegations were, quote, "holy out of character" and noted his client was charged only with violating a federal recording statute, not stealing trade secrets or acting as a foreign agent. >> Okay, so how did OpenAI ultimately decide that this bug was planted by Al-Fazan? >> They weren't sure at first, so that basically immediately prompted them to start trying to look around to see, you know, who was even in this office in the first place? Who could have put something like this here? So they didn't have cameras to actually look for who might have put it there, but they had cameras on the outside. So they could see who is kind of in the vicinity around that time. And the person that came up who was kind of walking around this office, you know, for several weeks leading up to this, who wasn't really supposed to be around there, was this 35-year-old engineering lead at the company Abdul Rahman Al-Fazan. [MUSIC] So one of the investigators at OpenAI decided to put a camera behind the TV. With the idea that whoever put this microphone here, at some point was going to come back looking for it to try to retrieve it. And what this produced is, you know, roughly a week after OpenAI discovered this device, is a suspect in question peering behind the TV. And they got this incredible photograph of him with the kind of a look of shock on his face, peering behind the TV to not only see a camera taking a photo of him, but also that the device he had placed there was no longer there. >> He did look pretty shocked in that photo. >> Yeah, so things moved pretty quickly after that photo was taken. OpenAI had quietly been investigating this without tipping off Al-Fazan. They had alerted the FBI. And they were actually coordinating pretty closely with the FBI on this case. They were handing over evidence that could be helpful for the DOJ's investigation into what was going on here. The microphone itself was pretty basic. You can get the same one on Amazon for about $14. What made it weird was the packaging. >> There's not quite consensus on why it was wrapped in toilet paper. Some people have alleged that it was kind of acting as almost like a sound guard on a microphone to kind of maybe there was air conditioning unit in the building that they wanted to kind of shield the wind from other people have suggested that it was a way to conceal it, although you could argue that a device might have actually looked less conspicuous if it wasn't wrapped in toilet paper. And then yeah, other people have said maybe he was trying to not get his fingerprints on it. So he had wrapped it in something. But this is something I've talked to multiple experts about, and no one is quite sure why it was this way. So not a James Bond level spy effort? >> Yeah, so I think something that onlookers have kind of remarked about this is that this doesn't appear to be the most sophisticated kind of spy attempt they've ever seen. I mean, he was using pretty low tech equipment. He was prompting the company's internal tools to get information about how to conduct this. He was scoping out the conference room for weeks, and according to allegations made by the prosecution, he also seems to have misplaced his backpack at some point, which had another recording device in it. >> Alphazon was fired and arrested shortly after he was caught on camera. According to the Justice Department, the case against Alphazon is about, quote, "safe guarding our nation's economic strength and national security." But that means exactly isn't completely clear yet. But Max says Alphazon's ties to Saudi Arabia are a big part of the case that prosecutors are building. >> A big part of this story is that Alphazon is a Saudi national, and what the DOJ cited in their complaint is that this is a national security concern, that this, you know, they pretty directly have invoked national security here. >> Alphazon also has connections to powerful people in Saudi Arabia, according to Max's reporting. >> Last year, when he joined OpenAI, around the same time, he also joined this venture capital firm called Raid Ventures, which is a Saudi Arabian-based venture capital firm. And another person who's a partner at the firm is a member of the Saudi royal family. >> According to prosecutors, Alphazon was also in regular communication with an unnamed member of the Saudi royal family. The royal family member told Alphazon that an artificial intelligence initiative was being launched in Saudi Arabia, and he wanted Alphazon to join it. >> There was a text thread between Alphazon and this member of the Saudi royal family that prosecutors showed in court that said that he was talking with this person as recently as October 5th, and you know, one of the texts they had in the last year said something along the lines of, "Saudi needs you more than the U.S." And Alpha-Zon responded 100. I'm flying that he agreed. But apparently, he also responded, according to the prosecutors, Alpha-Zon said he wouldn't leave until his stock and opening eye had vested. Interesting. That said, you know, as Alpha-Zon's lawyer has said many times, in this case, so far, you know, someone's nationality and whether born does not make them guilty of anything. I mean, I think some people see this as examples that he was very close with the Saudi Arabian community and parts of it. At a detention hearing this week, prosecutors framed Alpha-Zon as a flight risk. They highlighted the fact that he had purchased a one-way ticket to Saudi Arabia and was set to depart just three days after he was arrested. Alpha-Zon's lawyer says that he had booked a one-way ticket for him, his wife, and his child. He has a three-month old to Saudi Arabia. And I think, depending on how you look at it, people can read into that different ways. But, you know, Alpha-Zon has basically said, you know, his wife was dealing with immigration issues and they were headed back to Saudi Arabia. He was scheduled to start taking paternity leave. His lawyer said Alpha-Zon typically buys one-way tickets to save money. The judge expressed skepticism about that claim. The concerns about espionage don't start and end with Alpha-Zon. AI companies know the problem is only getting bigger and they're trying to fight back. That's next. OpenAI is so worried about spies. It's telling employees to be on a high alert in and out of the office. One thing that we've reported on is that, you know, OpenAI has given employees these trainings about how to ensure that they're not being targeted by a spy, being socially engineered. And one of the things they tell people in training is, hey, if you're a young man, you work at OpenAI, and a beautiful woman comes up to you and tries to talk to you, you know, what you should think about is this a spy. That's actually not interested in me as a young man, but as, you know, a person with access to OpenAI's information, which is kind of hilarious to think about the fact that, you know, if you're a person at OpenAI, you can't quite trust the romantic reactions that you're having. They've also worn their employees to watch out for pretty much anybody. Another kind of scenario that employees were presented as, you know, hey, you're at a playground with your kid and, you know, a stranger trying to befriend you as an OpenAI employee, is there with their child and they come up and they make small talk with you. And OpenAI is basically told employees, you know, you should be wary of this happens. I think it kind of speaks to, you know, the level of paranoia they're trying to instill in their employees about this information. Employees are told to use green protectors on public transit and to clear whiteboards when they're finished with them. Just in case kitchen and cleaning staff are compromised. They've even been warned not to wear branded clothing in public, so there might be fewer OpenAI quarter zips in the valley these days. So now let's zoom out a little bit. Why would someone want to spy on OpenAI in the first place? Yeah, so I think that OpenAI in particular, they are very secretive about the way they train AI models, the way they build products around their models, their interactions with government officials, because they think that all of those are trade secrets that basically coalesce into this secret sauce of why OpenAI is the most powerful and most important AI company in the world right now, except for maybe Anthropic. Corporate espionage isn't anything new, but Silicon Valley's concerns are twofold, losing trade secrets to competitors and losing trade secrets to foreign governments. American companies have long held secrets that foreign adversaries around the world would like to gain access to. There's so much fear of giving another country China an edge on America. This new class of AI models has heightened government fears around AI more broadly. Ever since Anthropic released this mythos model in the last year that set off alarm bells in Washington because of its abilities to basically hack into legacy software systems. I think there's a real fear right now that these AI models present a cybersecurity concern that we haven't quite seen before and that if a bad actor were to get their hands on one of these models, I think some government officials, including in the Trump administration, are concerned about giving another country access to that kind of technology. And are there any previous examples of that happening? In 2022, there was a case where Twitter employee was caught spying on the company on behalf of Saudi Arabia. I'm at Abu Amal handed over the personal details of critics of the Saudi government, including email addresses and phone numbers. In return, he received hundreds of thousands of dollars. He was basically collecting private user information about Saudi Arabian critics and accepting payments and sending that to the Saudi Arabian government. That Twitter employee was convicted and sentenced to 42 months in prison, although his case was overturned by the Supreme Court because he was tried in the wrong district. Beyond national security concerns, the stealing of corporate secrets is also just bad for business. As much as all of these companies are kind of representing America on the global stage, they're also fiercely competing with each other to get a leg up on each other in the AI race. I mean, they do have to protect their trade secrets. These companies are also trying to IPO right now. They're trying to navigate government regulators and get the right regulatory regime in place so that AI can thrive. The threats coming from within the industry have made their way to court. Apple sued open AI alleging trade secret theft. Open AI is started in the court filing. The lawsuit is baseless. The case hasn't been resolved yet. A federal judge wrote this week that Al-Fazan should be held without bail. Whether or not he's actually a national security threat will be litigated in the coming weeks. But his case and how quickly he was caught is already shedding light on how these companies are responding to such threats. I talked to an FBI official and they said it was actually a pretty impressive example of a private industry company and the public sector with the FBI working together to do something like this pretty quickly. This is not just a normal company here. I mean, this is open AI. So I think that there's kind of a couple things converging here to this point where even just the implication that a Saudi national was spying on one of America's leading tech companies, I think this is really kind of woken people up as another reminder of how sought after some of these secrets are. Max says these companies are walking an almost impossible tight rope. AI companies are simultaneously talking with almost everyone about the work they're doing. They have to talk with government regulators. They have to talk with other companies that they're selling their products to. They're trying to talk to third parties about auditing them. And at the same time, they have to be so secretive about some of the details of what they're working on. And it almost feels like this kind of like impossible space they have to operate in, where you know, they are trying to share a lot about what they're doing to kind of rally support and bring people in on the process and hire thousands of people to work on these new projects. And at the same time, be so paranoid about some of the details that, you know, they have to really highly scrutinize even their own employees. This episode has been updated. A previous version incorrectly said OpenAI countersued Apple in their trade secret dispute. That's all for today, Thursday, October 22nd. The journal is a co-production of Spotify and the Wall Street Journal. Additional reporting in this episode by Angel Algang, Megan Berbrowski, Georgia Wells, and Aaron Wu. Thanks for listening. See you tomorrow.