FBI Assistant Director Bryan Vorndran: Hello, everyone. Welcome back to Ahead of the Threat. Bryan Vorndran, Assistant Director of the FBI Cyber Division. With me, as always, is Jamil Farshchi, the Chief Technology Officer at Equifax. It's been a while since we've been with you, but we're happy to be back. And we're going to have a great episode with Scott Aaronson at the University of Texas, probably the leading researcher on all things quantum in the United States. And so, stay tuned for that after Jamil and I cover the Top Three.

Again, apologies for being away for a month or so, but I'm happy to report that, you know, from an FBI perspective, our commitment to victims of cyber intrusion remains unchanged, what it has always been and always will be, that's paramount, and in the FBI's DNA for 115 years, in terms of serving victims of crime. And in terms of our strategy, it's all about public and private sector relationships, both domestically and internationally, to impose cost on our victim—or to impose cost on our adversaries and slow them down. So glad to be back with you.

We're going to hit our Top Three here today, and I'll lay them out here and then we'll get into them. First is Google's acquisition of Wiz for $32 billion. The second is the AI Act in Europe, and last is Hellcats attack on Jaguar Land Rover. And we'll talk about that.

So, jumping right into Wiz, Wiz is all things cloud security, right? Very, very functional cloud security company. I believe Israeli-based but headquartered here in New York City, and $32 billion acquisition by Google continues to show the consolidation of the cybersecurity industry in terms of consolidation into the bigs, the strategic bigs. Jamil, your thoughts?

FBI Strategic Engagement Advisor Jamil Farshchi: I, well, first, congrats to the Wiz team. We've been using them for several years now. And it’s been a fantastic company and they've been great partners. They actually helped us. We had some unique requests about transparency because I wanted to be able to provide our cloud security posture scores out to our customers for the products and services that they leverage from us at Equifax. And they worked diligently to be able to build a nice, unique front end for us, just especially for Equifax to be able to help support that transparency within the industry.

And so, these guys have been great, really, really happy for the acquisition. Props to them, I think the initial offer, I think it was just this past summer was around $20 billion. And then it's now…they parlayed that into a $32 billion exit on this one.

I think the interesting thing going forward will be where…what do the other major cloud providers do? Because as you can imagine, with Wiz's customer base, a significant number of them are on AWS or on Azure, and so I'll be curious strategically and competitively what those, what those other players choose to do.

But I think this is a strong buy for Google. And it continues down the path that they've been on for a while. Remember, what was it, a year, year-and-a-half, two years ago, they purchased Mandiant as well. And so, I think this shows Google's commitment to cybersecurity. So, it'll be…I think it's a good acquisition.

Vorndran: Okay. Great. Well Jamil, I'm going to go to you for the second one, the Artificial Intelligence Act of Europe. What are your thoughts?

Farshchi: This is a…interesting. So, we all know about, you know, GDPR and I believe it's 4% fines of global revenue if you violate those rules. Up to 4%. This one on the AI side is even greater. It's 7% of global revenues if you're in violation. They’ve had these guidelines in place for a while now, but they just went into enforcement mode, as of…I think it was February 2nd. And so, I think it was Warby Parker or someone, like the company that makes the glasses, was, I think, the first organization that was fined here for an indicent that they had.

I think the interesting thing about this one will be how it ultimately plays out, because some of the rules are fairly nebulous. I mean, there's, you know, things like, you must have adequate AI literacy. Things like that are pretty tough to be able to quantify. And so, you know, some of the social scoring things or, you know, the biometric surveillance and stuff like that -- that'll be easier, I think, to be able to measure and stuff.

But I'm interested to see how it plays out. GDPR ended up being fairly meaningful. We haven't seen a tremendous number of fines coming out on that one. So, there was a lot of hype around it and a lot of fear; I think fear mongering. But it didn't really play—turn out to play out that way in the long run. This one could be different.

I mean, AI is advancing at an incredible, incredible pace right now. And so, keeping your arms around it from a regulatory standpoint is super, super challenging. And making sure to revise and update this stuff will be a tough challenge, I think, for the regulators. But it will also be really difficult for the companies that are required to comply with them because it's difficult to interpret exactly what you're supposed to do, especially in such a quickly evolving space.

But, you know, the line in the sand has been drawn. The enforcement actions are there now. And so, we'll see how it plays out. But, I think buckle up is probably the best guidance for everybody.

Vorndran: I just want to give the audience a few statistics. You know, to Jamil’s point about how quickly the artificial intelligence space is advancing. And just two statistics to really bring that home. Number one, within a year, 90% of all code will likely be written by machines instead of humans. And within one year, machines will out code, in terms of quality of code, the best human coders on the planet, right?

Farshchi: Hang on, hang on. I'm sorry. I have to cut in here.

Vorndran: Please do.

Farshchi: There is so much hype in the AI space. And look, I'm a huge advocate and we're leveraging it and experimenting in a lot of different areas. These…those two data points, you just cited are from people that have a very strong self-interest in AI truly doing that. But that…to believe that 90…I think it was the Anthropic CEO that said that if I remember right? 90% of the code that we create is going to be developed by AI by the end of the year is… no way, is that ever going—

Vorndran: I didn't say by the end of the year, I said within a year.

Farshchi: Within a year. Sorry, sorry. By the— I mean, okay, whatever, so within the next 12 months, 90% of the code, I just, it's not going…there's no way that is going to happen. Do—are companies going to substantially increase the amount of code that is developed by AI through codices and stuff like that? Absolutely. There's no question; we're doing the same thing.

But what you find with this is that for the most part, the code that is being developed; that is effectively developed by AI today is really low-level code. It's really basic stuff. And so, it does help tremendously from an efficiency standpoint. But the likelihood that you get into complex code with AI today or even in the next year seems really, really unlikely. And it's not very good today at all, at being able to refactor preexisting code. And so, I think there's a lot of hype here. I think these claims are very, very hypie. And so, we'll see where it goes. But you never know.

Vorndran: So let me ask you a question then, if we'll do a little hand-to-hand combat here.

Farshchi: Bring it.

Vorndran: What are your thoughts on the computer science space in terms of education and what the value is of that in 10, 12, 14 years in terms of degrees and skillsets?

Farshchi: I think it's going to be… So, I got this question at an all hands the other day. One of the developers on my team, he was really, genuinely worried. And I completely empathize with him, too, given all the headlines and stuff. But he's like, ‘Jamil, am I going to lose my job? Like, I'm super passionate, I love developing, but with all of this AI stuff, I don't know that there's going to be a job for me going forward.’

But there—I'm…as long as you consistently develop your skillsets and continue to improve, it is hard for me to imagine a world where—in the near future at all—where these folks will not have a job to be had. I mean, if you look at technology and it's happened consistently for the last, I mean, at least the last 20 years, where there's always some new tech that comes along and, you know, a new IDE (integrated development environment) or new language or whatever it might be. And then everyone's like, ‘oh my gosh, this is going to completely change the game.’ And developers are going to be, you know, we're not going to need them anymore. I think the same thing applies here.

Now, do people need to level up? And is there going to be a level of sophistication that you're going to have to meet to be able to do it? Or move into more of a supervision-type of capacity? Dive into being really good at, I don’t know, vibe coding and stuff like that? Sure, but we all have to adapt and evolve.

And I think from an education standpoint, we just need to be thoughtful around the programs and what we're teaching for…what we're teaching the kids to make sure that it aligns with where the future is going. And the schools, the education system is not, quite frankly, done a great job of that as of late. I mean, a lot of the interns and the development programs that for kids that we bring in from college, they didn't even learn about cloud. Like, it’s mind blowing that today, in this day and age, like you do not understand the fundamentals of cloud. You don't learn that in college. And they're learning it as they do internships and stuff at Equifax. So, I think that the education system and the universities in particular need to evolve and adapt to the latest stuff. If they don't, then it'll put everyone at a disadvantage.

But going back to the original point, I just do not see a world where vast numbers of developers are without work because AI has just completely replaced them. Of course, it'll happen at some level, and I think AI will have a demonstrable effect on a lot of other roles out there that it will be able to scale and have a better level of accuracy on some of those kinds of things. But on the developer side, I'm just not seeing it right now.

Vorndran: Now, we will check back in late March 2026 and see where we are. All right.

And the last one… you know, on our comments here on Ahead of the Threat, we've been talking a lot about supply chain, third-party risk. And I learned of a new one just last week in terms of one portal that's used by all publicly traded companies. And I'm not going to share that here, but it just does go to show the consolidated nature of if an adversary wants to have an outsized impact, third party hits, supply chain hits are very, very significant ways to accomplish that.

So, Hellcat ransomware group hit Jaguar and Land Rover, this just in the past couple of days, through credential theft at LG. So, another supply chain attack on a major company that we all know. Jamil, your thoughts?

Farshchi: Yeah. This is, you know, it's interesting because a lot of times folks don't think about it. They think about supply chain, and they think about, oh, we got to vet our vendors and stuff like that. But a lot of times folks don't think about the credential risk and how, you know, we oftentimes not just share data and transmit data amongst different parties or store our data there, but we also, in many cases, provide access, via tools and even direct connect, to our environments, to third parties with these…with obviously with credentials.
And so, you look at this situation, and this is a third-party set of credentials that were compromised and then used to be able to breach this organization. And it's just this happens like there's hundreds and hundreds and hundreds of these exact same kinds of things. And so, when you're looking at your identity and access management and you're trying to protect them against the threats, it's not just access and credentials for your enterprise itself.

You've got to look at the third-party credentials as well, because this has happened for years now, and this is just yet another example. And it becomes easier and easier as that trove of credentials, compromised credentials, expands for bad actors to be able to take advantage of them and quite frankly, do their attack, execute their attacks at scale. You get one vendor that everyone uses. You get that trove of creds and then you can just hit any of the organizations that are leveraging these widely, widely adopted toolsets.

So beware, but don't scope down your supply chain security to just who's got direct access into your environment. You've got to look at all of the tools, particularly the SaaS ones, that you're leveraging as well.

Vorndran: Yeah, this has become an area of tremendous interest and passion of mine. I don't know how to fix it. But one of my goals has become to add value to that broader problem set over the coming years. And hopefully something I can do.

Farshchi: I think, Bryan, the longer term solve here has got to be the and I think we've talked about it before, even maybe with Kevin Mandia a few months ago, the…this…the longer term solve has got to be to get rid of these static credentials. I mean, this is a methodology, a control, that we have used since the dawn of computing, basically. And yet it's still persists to this day.

And there are new technologies that solve for this problem. We just need to get to a point where people are able to readily utilize them. And we need to feel comfortable that, hey, look, we can actually eliminate this threat entirely. I mean, think about it: how many, attack vectors do organizations have and tools that we have as leaders to defend against them, where you can fundamentally eliminate the entire vector?

Kevin said it himself. He's like, ‘hey, EDR, enterprise detection response.’ Those tools have gotten so sophisticated nowadays. I think Crowd Strike even said the other day, too. ‘Hey, it's…we're so good at this nowadays that the attackers are moving to the next weakest link and that weakest link is identities, which is underpinned by static credentials: your usernames and passwords.’

So, if we have the ability to be able to get rid of them entirely, that vector’s gone. And so, I think, you know, we did this last year at Equifax. We eliminated them. I think it's the new wave. And when I think about it from an ROI, security risk ROI standpoint, it's the number one thing that we should all be trying to focus on.

Vorndran: All right, great. And just for the audience, the FBI and CISA just published a Medusa Ransomware joint cybersecurity advisory. We'll post it in the comment section below the YouTube stream here for anybody who wants to review it.

So that ends our Top Three for today. We're now going to go to a previously recorded episode with Scott Aaronson. Again, Scott is the leading expert on all things quantum. He's a professor at UT Austin. Great conversation with him. It does get a little deep at times, in terms of technology and math. But ask the audience to stick with it because at the end we talk about what types of cryptographic, you know, whether it's AES, or RSA, etc. actually in play to be broken by quantum computing.

So, with that, have a great day and we'll go to Scott.

***

Vorndran: Well, as I mentioned, joining us today on Ahead of the Threat is Scott Aaronson, University of Texas at Austin and a lead researcher on quantum. Scott, welcome to the show. Can you give our audience a brief background on yourself?

Professor Scott Aaronson, University of Texas at Austin: Yeah, thanks. It's an honor to be here. So, I'm a theoretical computer scientist. So I've spent about 25 years, mostly studying the capabilities and limits of quantum computers. What we could or couldn't do with computers that we don't yet have. So, you know, I don't build devices. I, you know, don't have, a lab with qubits or you know, dilution refrigerator or anything like that. You know, we've got a fridge for leftovers.

But we… you know, basically do math. We try to prove things about quantum algorithms, you know, and about how quantum mechanics would change the theory of computation. So that's been most of my career. Now I'm just getting back from a leave of absence; I took two years off to work at OpenAI, in their alignment team, which, unfortunately, doesn't exist anymore. But we were thinking about, you know, how to use theoretical computer science to make AI safer. And I thought about things like, what are marking the outputs of large language models like ChatGPT?

So, you know, now I'm back in academia and so trying to get back into quantum computing while also still thinking about AI because how can anyone ignore it right now?

Vorndran: Jamil, you want to go or you want me to go? I got a ton of questions.

Farshchi: Yeah, I'll start. So, Scott, thanks for being here.

Aaronson: Sure

Farshchi: As an Oklahoma Sooners undergrad, I will, make sure that this is not an adversarial discussion, with you as you.

Aaronson: All right. I have to, you know, hook ‘em! Yeah!

Farshchi: Look, can you…you've got a fantastic background here. Can you sort of just give us the lay of the land at a high level in terms of where quantum is? We've been… people have been talking about it for ages. It seems as if, you know, there's all kinds of different predictions about whether it's going to come to fruition in the next three years, seven years, 20 years. Just sort of tell our audience about where things are today?

Aaronson: I mean, you know, I could spend, you know, days, months, years, you know, answering questions about quantum computing and yet still would not have an answer to the question that everyone wants to know—how many years will it take? Right? That's just not something that theory lets us answer. But I—what I can tell you, at least, is about where we are now and how we got here and what still needs to be done.

So, you know, the basic ideas of the theory of quantum computing, you know, as we still have them today were mostly developed in the mid-1990s. Okay, that's when you saw Shor's famous factoring algorithm, which was, you know, the first really clear demonstration that a quantum computer could indeed give you a huge speed-up over a classical computer for solving a problem of practical importance you know, that's got nothing to do with quantum mechanics, you know, namely, factoring numbers and thereby breaking the encryption that protects most of the internet, right?

And so then, you know, that really motivated people to think about this, you know, not just as an idea which, you know, had been kicking around for a decade or more at that point. But as how do we actually build this? You know, what would it actually take? And, within a few years after that, after Shor's algorithm, we had a whole theory of quantum error correction, okay? That basically said here is what it would take, right? You don't have to get your basic components, your quantum bits, your qubits, you know, to be infinitely reliable. Okay?

But you do have to get them to be very, very reliable, right? Once they pass a certain threshold in reliability, then there were these very clever, quantum-error correcting codes that can get you the rest of the way. And it can then, you know, use your sort of noisy qubits to effectively simulate qubits with less and less noise.

And so, then you can effectively keep a quantum computation going for as long as you would need it to go. And so that’s sort of set the engineering goal of the field for the next three decades. You know, that discovery of the theory of quantum error correction and quantum fault tolerance. Okay? Then, you know, I think most physicists and computer scientists were convinced at that point that, ‘okay, there is no new physics that is needed in some sense,’ Right? There's no…you know…the basic physical principles, you know, of quantum mechanics that we've had since the 1920s, you know, since Schrodinger and Heisenberg and those guys, you know, will be enough, you know, if those work like the textbooks say that they do, then yes, it should be possible—in principle–to do this.

But, you know, it will be a staggeringly hard engineering problem, right? And I mean, I try to, impress, you know, the scale of this on people by pointing out that, you know, in the 1820s, you know, Charles Babbage, right? Already had, you know, the basic ideas of what we would today call a programable classical computer, right? But he couldn't build it. You know, the technology just wasn't ready for it. And it took more than a century for the transistor, you know, and the integrated circuit to be invented that really, you know, made his dream practical. Okay? And, you know, I'm hoping that it will be less than a century in this case. Right?

But, I mean, you know, it's been, let's say 30 years or so since, you know, quantum error correction was discovered. And what's very exciting is that, you know, we're not there yet, but there is unbelievable progress that has been made on the experimental side. And, you know, it's a particularly exciting time right now.

I mean, just within the last year, we have seen, you know, some of the things that were theorized in the 1990s, you know, finally demonstrated experimentally. In particular, Google just, in December, you know, announced this Willow chip, which actually demonstrates an encoded logical qubit; sort of an error-corrected qubit. Where, you know, the key threshold that they've crossed is that as they scale to larger and larger system sizes. So as they use more and more physical qubits to encode their logical qubit, the logical qubit stays alive for longer and longer, rather than shorter and shorter. Right?

So, you know, that's just one logical qubit. You know, that's a long way from thousands of logical qubits that will, you know, maintain their quantum state for possibly hours or days, you know, in order to break a cryptographic code, or, you know, simulate a high temperature superconductor or, you know, the other exciting things that people want to do.

But I think, you know, if you look at where the experimentalists are now compared to where they were, you know, 20 years ago, 10 years ago, five years ago… I mean, you know, you think… holy cow! Right? This is, if this is given enough time, you know, this is actually going to work, right?

There's, you know, of course, there are enormous questions. You know, will the funding run out? Will people just lose interest? Will civilization collapse? You know? I sometimes worry about that, you know. Will AI, you know, eat up everyone's interest and attention and just make this irrelevant, right?

So those are, you know, and of course, none of this addresses the other question of what is a quantum computer good for, right? And there, you know, unfortunately, you know, the story is not the one that a lot of people want to hear. You know, the applications are more specialized than a lot of people would like them to be. Okay. But, you know, will people just decide, oh, you know, maybe this isn't this isn't worth, you know, what it would take to actually build this?

So, you know, those are all questions. But I think if people want this enough, then, you know, especially given the experimental progress over the last few years, I think that, it seems to me like, you know, a question of time until they can have it.

Vorndran: So, Scott, let's build off of your… what you just mentioned about the, you know, you mentioned a computer doing this or what a computer looks like that does this… but functionally and practically, what is quantum and why is it important for the evolution of our society from your perspective?

Aaronson: Well, okay, the evolution of society is a hard thing for me to address, right? We don't know how, you know, new technologies will change society. But what I can certainly tell you about is, you know, what is quantum mechanics, right? You know, why is it interesting? Why is it different?

So, you know, people often, you know, have heard something that, you know, quantum mechanics is complicated, confusing… You know, Einstein, you know, didn't… never liked it. that said, you know, “God,” you know, “does not play dice.” And, you know, that Niels Bohr said to him, you know, “Einstein, stop telling God what to do.”

And so, you know, there's all this sort of cultural baggage around it. Okay. And, indeed, you know, if you want to apply quantum mechanics to actually understanding the behavior of atoms and molecules and, you know, photons and electrons, as the physicists do, then, you know, that's pretty complicated. There's, you know, lots of years and years of things that you need to learn. Okay?

But, the way that we think about quantum mechanics and computer science is, you know, the way that that's enough for quantum computing is actually much simpler than that. I like to say, you know, quantum mechanics is surprisingly simple once you take the physics out of it. And so we think about it as just a certain generalization of the rules of probability.

So, you know, we use probability all the time, you know, even in classical life, right? You might say there's a 30% chance of rain tomorrow or of, you know, school being closed for snow or, you know, this candidate winning an election. You would never say that there's a -30% chance, right? That would just be nonsense. Okay. Even less, would you say that there's, an “I” chance, you know, a square root of negative one chance? You know, what would that mean?

Okay, so the key thing that quantum mechanics says about the world is, well, you know, first of all, that there are probabilities, you know, at the most fundamental level of physics that we are able to probe, right? Like, you can have… shoot a photon at a screen, do it multiple times, exactly the same initial condition, you know, exactly the same state of the photon. And sometimes it will go one way, and sometimes it will go another way. Right? And, there is, you know, according to our current understanding of physics, no way to predict which way the individual photon will go. You can only calculate the probability, right?

And so, you know, many people have heard that. Okay. But I'm here to tell you that that that itself is not the weird part, right? If it was just a matter of probabilities, then, you know, we could always reconcile ourselves to that. We could always say, well, probably, you know, ‘every photon has just the secret little RFID tag on it that, you know, we don't get to see that secretly determines which way it's going to go.’ Right?

And, you know, even in high school chemistry, right? When the teacher said, well, you know, ‘physicists used to think that the electron orbits the nucleus,’ you know, like a little planet orbiting the sun. But then they discovered that it's, you know, not in any one place, and it's just in a big smear, all around the nucleus.

And I said, ‘that sounds like nonsense.’ You know, surely that's just a fancy way of saying they don't know where the electron is. Right? But okay, if all that was going on was probability, then we could just say, you know, ‘there's always just some more details that we don't know.’ Right? And if we knew those details, then everything would be deterministic. Okay?

But quantum mechanics is different from that, okay? Because it what it says is, you have to change the way that you calculate the probability that something happens. Okay? And that's, you know, you wouldn't even think of that as physics. That seems like math. Okay? And yet, you know, 100 years ago, physics came along and said, ‘no, there's a different set of rules for probability.’ Okay. And these new rules involve numbers that are called amplitudes.

Okay. So amplitudes are related to probabilities. But they don't have to be between 0 and 1. Okay? They can be positive or negative. And in fact, they can be complex numbers, you know, involving the square root of minus one. And so, then what quantum mechanics says it's basically that says like ‘every possible way that a physical system could be.’

So, for example, every place where electrons could be found, you know, if you measure them has some amplitude associated to. Right? So, the truth—the actual truth—about what is going on with, you know, a bunch of electrons in a molecule is just this wave of amplitudes, right? That is the real state of it. It's an amplitude for every possible configuration of the electrons. Okay?

And now if you want to know how likely it will be to see the electrons in some place when you look, you have to take the amplitude for that configuration, and then you take its squared absolute value. That gives you a probability; it gives you a number from 0 to 1. Okay? So there's a rule for turning these amplitudes into ordinary probabilities. You know, one of the most famous rules in physics, it's called, the Born Rule. Okay. But now if you're not looking, okay? If you keep your system isolated, if you don't measure it, then these amplitudes can evolve in time by their own rules. Okay. Which are different from the normal rules of probability.
And the main way that they're different is that these amplitudes, you know, being complex numbers, can, as we say, interfere with each other. And they can cancel each other out. Okay? So what does that mean? It means, like if I have a photon, let's say, and it can reach a certain spot, by taking one path, with a… but with a positive amplitude. Or it could reach that same spot by taking a different path, but that other path has a negative amplitude. Right? But now the total amplitude for the photon to reach that spot will be the sum of this positive and the negative contribution. Right? And if those are equal in magnitude, then they can cancel each other out, which means that the total amplitude is zero. And then the photon never arrives at that spot at all. Okay?

Whereas if I were to block one of the two paths. So, you know, and this is what happens in the famous two slit experiment. But, you know, Richard Feynman used to say that all of quantum mechanics is contained in this one experiment okay.

Then, if you, so you shoot a photon at the screen with two slits in it, you know, giving it two paths to travel through. And then you see that there's these certain spots where the photon never shows up. But now if I close one of the slits, which means if I block one of the two paths that the photon could take, then I only get a positive contribution or only a negative contribution, which means that now the photon can appear there. Okay, now the amplitude is not zero.

So to say that again, by decreasing the number of paths that a photon can take to get somewhere, you can increase the chance that it gets there. Okay. So that that's the sign that we're dealing with new laws of probability. And, you know, this is maybe the most dramatic thing that happened, you know, in the history of physics, you know, since Newton. Right? It was this change to the rules of probability. Everything else that people say about quantum mechanics, you know, about entanglement and about tunneling and, you know, on and on, you know, and so in some sense, they're all just logical consequences of this one change to how we calculate probabilities.

Vorndran: That's sounds a little complicated, Scott.

Aaronson: Well, I mean, okay. But, you know, I mean, it takes 10 minutes to explain it, right?

Vorndran: Yeah.

Aaronson: You know, it doesn't take a year to explain it. Right? It's, you know, and in, you know, 20 years of journalists asking me to explain it in one sentence, you know, that that's about as far as I've been able to compress it.

Okay. But a… you know, now a key point is that, you know, as you add more and more particles, right? The rules of quantum mechanics are unequivocal that you need an amplitude for every possible configuration of all the particles together. Okay?

So if you had, let's say 100 bits. Okay. So things that can be 0 or 1. Okay. But now these are quantum bits, meaning they can have an amplitude to be zero and an amplitude to be one. Right? So, you know, the basic building block in quantum computing is what we call a qubit, which is just a bit that can be in, you know, a superposition, as we say, if a zero state and a one state, which means it's got some amplitude for each.

Farshchi: It can be in both at once though, can't it?

Aaronson: Sorry? Well, again, you know, people are always trying to round this down to sort of ordinary language. They say, you know, it's both of them at once, or it's either one or the other. It's a linear combination of the two. Right. That's, you know, there's no there's no better way to say it in English than that. Right? It's… when you look, you only see one of them. Okay. When you make a measurement, you see that it's zero or that it's one.

Okay. If… so, for examples… concretely, you know, your qubit might be, the spin of an atomic nucleus, like, is it spinning clockwise or counterclockwise about some axis? Right. When I measure when I look, I'm going to see it's always going to be one or the other clockwise or counterclockwise, okay?

But if I want to calculate the probability that it's one or the other, then I need to use these amplitudes, okay? And there's an amplitude for the clockwise possibility. There's an amplitude for the counterclockwise possibility. Right. And quantum mechanics would say that, you know there is no deeper truth of the matter than that. Okay. If I have amplitudes for multiple possibilities, this is what I call a superposition. Okay. A qubit is just a bit that can be in a superposition of the of the zero state and the one state. Okay.

But now, where it gets even more interesting is when I've got multiple qubits. Okay? Because if I've got two qubits now, there's four possibilities. There's 0-0, 0-1, 1-0, and 1-1 okay? And each of those gets its own amplitude, okay?

If I've got three qubits now there are eight possibilities right. If there's 10 qubits that's 1,024, right? And if there's a thousand qubits now, you know, which is not that many, right? You know, a thousand bits is not a lot in a modern computer, okay? But now I've got two to the thousand power amplitudes, okay?

Two to the thousand power is more than a Google, okay? It's more than the number of electrons that, you know, that there are in the observable universe. So, you know, so in some sense, quantum mechanics has been telling us for 100 years that, you know, just to keep track of the state of a thousand particles, you know, say is, sitting somewhere, nature has to maintain some scratch paper with two to the thousand complex numbers, right? Two to the thousand parameters. And every time something happens to those particles, nature has to cross off all those numbers and replace them by new numbers, right? Now that's a staggering amount of work for nature to be going to.

And, you know, the chemists and physicists have known this for generations, right? They've known it mostly as a practical problem that, you know, if you're trying to simulate quantum mechanics with a classical computer, for example, because you want to understand, you know, how will this molecule behave? You know? How will, you know…will it bind to a receptor? You know, like a drug that I want, right? Or will this material, you know, conduct electricity or not, okay? If you want to do those calculations, you know, you have to keep track of this exponential number of amplitudes, right? And that's a staggering problem.

And a lot of what chemists and physicists have been doing for generations has been inventing tricks and approximation methods to deal with that exponentiality. Okay, the best tricks, like density functional theory, were awarded with Nobel Prizes. Okay. But, it was only in the early 1980s that a few physicists, like Richard Feynman and David Deutsch, had this idea that if nature is giving us this computational lemon, you know, all these amplitudes to keep track of, then why don't we make lemonade out of it? Right? So why don't we build a computer that would itself work on these principles of, you know, superposition, interference of amplitudes, you know, this, this sort of exponentially large state space,? And, you know, they called that a quantum computer, okay?

Now, of course, they then face the question: Well, supposing that you build a quantum computer, what would it be good for? And, fortysomething years ago, they only had one answer to that question, which is, well, it would be good for simulating quantum mechanics itself. Okay. And, you know, I think that, you know, despite everything that's happened in the ensuing decades, right, that remains the most important economic application for quantum computers that we know, right?

I think, you know, simulating materials and chemistry, you know, that's actually the thing where we're… the place where we're most confident that a quantum computer could provide economic value to the world. Okay? Even today, okay?

But, that was not what got everyone excited about it. As I said before, the thing that really got people's attention was this discovery in…what came later in the 1990s, that a quantum computer can sometimes get a huge speed-up even for solving a purely classical problem that's got nothing to do with quantum mechanics, right? And the famous example there is Shor's algorithm for factoring numbers, okay?

But now we really come to sort of the central misconception that people have about this subject, which is that people say, well, because it's, you know, people learn that this sounds good, right? They said, well, a quantum computer is just a massively parallel computer. It just tries every possible answer in parallel. And that's how it gets the speed up, right? So, you know, on that view, Shor's algorithm, for example, would work by just trying every possible divisor in superposition. You know, sometimes people say each one in a different parallel universe, okay? And then, you know, somehow, you just magically pick the best one you met. You pick the needle in the haystack.

Well, if that was how it worked, then that would be, you know, that would be useful way, way, way beyond this specific problem of factoring numbers, okay? But unfortunately, that's not how it works. Okay.

And the core problem comes back to what I said before, which is that when you look at a quantum state, you force it to decide which outcome it's going to give you, and it just picks one randomly. And whichever one it picks, it just sticks with that. Okay. So, it's true that you could create a superposition over all the possible answers to your problem, even if there were exponentially many of them.

Okay. But for a computer to be useful at some point, you have to look at it. You have to measure, you have to get an output, right? And if you just measure a super… this equal superposition over all the answers, you know, not having done anything else to it, the rules of quantum mechanics tell you what you'll see is a random answer. And, well, if you just wanted a random answer, you could have flipped a coin. Or, you know, you could have just, saved a lot of effort and, you know, building this quantum computer, right? And so the only hope of getting an advantage from a quantum computer is to exploit the way that these amplitudes, you know, being complex numbers, work differently from probabilities. Okay.

So with every algorithm for a quantum computer, you know, including Shor's factoring algorithm, the trick then we're trying to do is to choreograph a pattern of interference. So that for each wrong answer, each answer that we don't want, some of the contributions to its amplitude are positive and others are negative. So on the whole, they cancel each other out.

Whereas for the right answer, the answer we do want, we want all the contributions to its amplitude to be pointing the same way. Okay, so they add up constructively, okay? And if we can do that, then when we measure, we'll see the right answer with a large probability, okay? But the hard part is, you know, you need to boost the amplitude of the right answer, even though you know, you yourself don't know in advance which answer is the right one. You know, if you already knew, what would be the point, right? And you have to do all of this faster than any classical, you know, even the cleverest classical algorithm could do the same thing because, again, you know, otherwise why not just use a classical computer?

Okay, so this is what Shor showed how to do. But to do it, he had to take advantage of very, very special properties of a few problems, like factoring numbers and some related problems in number theory that are very important in cryptography. And, you know, for 30 years since then, we've been trying to figure out, you know, how far can that be generalized to give quantum speed-ups for other problems, you know, including, you know, problems of, you know, more importance and, let's say, business or a practical, you know, scientific things, you know, beyond just breaking people's cryptographic codes and…yeah, and reading their email, right? And we've made some progress there, but, you know, quantum speed-ups, you know, remain more specialized than many people would like, because they all depend on choreographing these interference patterns.

So, you know, like, this is not just a faster kind of computer. It is not just a, you know, like a standard computer, but massively parallel. It is this new weird way of exploiting nature to do computation that, you know, I think no science fiction writer would have even had the imagination to invent.

Farshchi: So are you think…do you think that all of the, I mean—look within the security community, this has been, you know, one of the top fear factors, I think, that we've all had for several years now, which is, ‘hey, this is going to break the internet and it's going to make all of our jobs effectively impossible, and we're going to make all this investment, all other this stuff.’ Are you you suggesting that those fears are overblown. Or I mean…

Aaronson: I think.

Farshchi: They might be underblown?

Aaronson: Okay. So certainly for people who care about, you know, protecting their sensitive data, for protecting their encryption, there is a real fear there, right? I wouldn't say that it's like an existential fear, like, you know, a runaway climate change or AI destroying the world. I would say it's more of like a Y2K-type of fear, right?

Farschi: Oh, so it is a fear.

Aaronson: It is a fear for which we already know good solutions. You know, we've already, you know, in fact, you know, in principle, we already have solutions. And it's a matter of deploying those solutions.

Vorndran: What are those solutions, Scott?

Aaronson: So yeah, I'll tell you… so basically what Shor discovered, you know, 30 years ago, was that a quantum computer, could break, essentially all of the public key encryption that is the most convenient kind that people use to protect the internet. So that includes RSA, which is based on, the belief that factoring is hard, Diffie-Hellman, which is, based on the…

Vorndran: Taking me back, Scott, here to my InfoSec training. Keep going.

Aaronson: Okay. Yeah. Yeah, sure. Okay. Good, good. So Diffie-Hellman…

Vorndran: AES, I assume?

Aaronson: Okay. So not AES.

Vorndran: Oh, okay.

Aaronson: So for…okay…so…there are these specific cryptographic systems based on problems in number theory, okay? So RSA, Diffie-Hellman, and elliptic curve cryptography are the three biggies, okay. And those can all be broken if you have a quantum computer, right? And these encompass, you know, most of the public key encryption that we use, okay.

So any time your web browser is using HTTPS, you know, it is using one of these systems to encrypt the messages, right? Bitcoin is using elliptic curve cryptography for the signatures, okay. As is Ethereum, I believe. Okay. So, you know, so these are hugely important, okay. But, you know, even within the cryptography that we use today, they're not everything. Right? There is also symmetric key cryptography. There are cryptographic hash functions. There are things that are that are not based on these very specific number theory problems. Okay. And for those other things, we only know how to get a more modest advantage with a quantum computer.

Okay. So there's a—after Shor's algorithm, you can do the second most important quantum algorithm is called Grover's algorithm. Okay. Discovered in 1996. And this is an algorithm that can take any problem that involves searching through a list of, say, N possible solutions. And it can solve it using a number of steps that only grows like the square root of N. Okay, so compared to Shor's algorithm, Grover's algorithm has enormously wider range of application. Right? It can be applied to all kinds of practical optimization problems, search problems, you know, AI, machine learning problems. Okay? And, you know, breaking like completely generic cryptosystems, you know, not just the special number theory ones. Okay?

But the disadvantage of Grover is that the speed up is much more modest, okay? You go from N the square root of N, right. So, you know, if N is astronomical, like if you, if you would have to search through two to the thousand keys, you know, with a classical computer with a square root of two to the thousand, there's two to the 500, right? It's smaller, but it's still quite big. Right? So it's not turning an exponential into a polynomial the way that Shor's algorithm does. Right.

Factoring, you know, the best, to factor an N digit number, the best methods that we know with classical computers take time that grows exponentially with the cube root of N, actually, but at any rate, exponentially. Okay. And, so, at least that's the best that's publicly known. You know, if the NSA knows something better then they haven't told us.

But, Shor's algorithm factors in N digit number using a number of steps that grows only like N squared roughly. Okay. So that’s what we call an exponential improvement. And so for RSA, Diffie-Hellman elliptic curve crypto, you know, there would basically be a complete break if you have really practical, scalable quantum computers.

For things like AES, DES and so forth there is a Grover speed up, which means you just have to increase the key size, you know, maybe double the key size, and then you'll have about the same level of security as you had before.

Vorndran: Yeah. Scott, let me just jump in here with another question. Is it—and I don't want to get sidetracked on symmetric versus asymmetric crypto, right?

Aaronson: Yeah, but that’s part of it.

Vorndran: That’s helpful, right?

Aaronson: So for symmetric cryptography you know we've understood for a very long time how to base it on these sort of very generic hard problems for which we only expect a Grover speedup from a quantum computer, and not more than that.

Vorndran: That's actually helpful to hear.

Aaronson: Yeah, for asymmetric cryptography, you know, that's the harder thing, right? It's like the more magic you're asking your cryptographic code to give you, right, including, you know, the magic of public key encryption, you know, not having to agree on a secret key in advance, right? Which is sort of helped make the whole modern internet possible.

But, you know, that's, a sort of very special thing, right? That, you know, we mostly know how to achieve using these very special number theory problems. Right. And the danger in cryptography is always like the more structure you put into the mathematical problem that you build your cryptosystem around, right, I mean, the more you can do with that problem, right, by exploiting that structure. But also the more an adversary could exploit the same structure to solve the problem, right? And even if not with a classical computer, then maybe with a quantum computer. Right. So that's exactly what happened with RSA and Diffie-Hellman.

Vorndran: Okay.

Aaronson: But there's another important part of the story, which is that over the last few decades, people have succeeded in developing what we call “post-quantum” or “quantum-resistant” asymmetric key cryptosystems.

Okay. So, we now have public key cryptosystems that are based on problems involving high dimensional lattices, for example. Okay. That we're at least we can say that no one knows how to break these public key cryptosystems even using a quantum computer.

Farshchi: And those were released what, last year by NIST, think? (National Institute of Standards & Technology)

Aaronson: I think so, yeah. So NIST. NIST actually held a public competition from 2017 until 2022 to agree on standards for these post-quantum public key cryptosystems and signature schemes. And…you know, there were some contenders, you know, that had looked pretty strong before, but that were actually eliminated, that were broken as part of that. And I think the community really converged around these lattice-based systems. So it was a very useful process in that way.

And now, NIST is, you know, is urging everyone to migrate to these new systems. And actually, just, you know, even the NSA is speaking in public about it. You know, it was just at a conference with someone from NSA who was, you know, trying to spearhead the government's, the U.S. federal government's, transition to post-quantum encryption and what he said and, you know, and, apparently this is all public now, is that by 2031, they’re hoping that every federal agency will have the capability to use these post-quantum cryptosystems. And then I think a few years…that a few years after that they would actually be using them.

Farshchi: You say 2031?

Aaronson: Yeah, that's what he said. So…you know, there were all sorts of, you know, milestones along the way, you know, that involve, you know, so all sorts of compliance requirements and, you know, things that I don't, you guys would probably understand better than me. But so you know, hopefully that will be fast enough.

I mean, and, you know, you could say, you know, if all goes well, then we'll just all upgrade the way that we do cryptography and then quantum computers, you know, will come along. They'll scale. And, you know, at least for cryptography, it will mostly just be a nothingburger. You know.

Vorndran: I'm going to counter that here in this. And we're running out of time. So maybe I make one comment, Jamil, then I’ll go to you.

The concern from our perspective, Bureau, Intelligence Community, is that let's just say 2031 is good, right? The concern becomes, what has the Chinese… what if the Chinese already stolen, right?

Aaronson: Yes. Absolutely.

Vorndran: That under those, you know, Diffie-Hellman or ellipt— curve that they can therefore decrypt and offers them an advantage.

But Jamil, I'll go to you for a final question. Just because we are almost out of time here.

Farshchi: That was a good point there, Bryan, though, I mean, there's evidence were there you know, other nation states are already collecting the data with the expectation that they’re going to be able to decrypt it within the near future. You have to assume that, like any, intelligence agency is, you know, vacuuming up whatever encrypted data it can in the hope of being able to decrypt it later, perhaps using a quantum computer. And, you know, I say to people like, if you have secrets that, you know, need to stay secret for the next decade, then, you know, you should probably already be using post-quantum encryption, which, you know, I think Google and other companies have already started deploying.

Aaronson: You know, I'm very lucky that I don't have secrets, but, people who do should possibly already worry about this, depending on, you know, just how long they need the secrets to stay for.

Farshchi: Well, I mean, I think for me, it's encouraging to know that, you know, there's obviously solutions that are available today, and it's really just a matter of, you know, operationally and practically rolling them out. I do think that that hurdle is massive. This is not like a typical problem.

Aaronson: Absolutely.

Farshchi: ‘Oh, I can just, you know, make some investments internally within my organization or wherever. And I'm good. Yeah.’ Because it relies on the broader ecosystem as well. But for my…

Aaronson: I mean it's one thing to, just as a theorist, say, ‘Oh, this is the solution.’ It's another thing to actually roll this out and change, you know, every router, every web server, every web browser. So, yeah.

But you know, but I mean, you know, one question people have is if this is going to happen, we’ll all upgrade to post-quantum encryption and then quantum computers won't be able to break it. Then what was the point? You know, why even build the quantum computer at all? And you know, that's where I think quantum simulation is so important. You know, that will remain as a big economic application of quantum computers.

Farshchi: Yeah. I think there's a lot of opportunity outside of just breaking encryption. We're running out of time here. One last question. And I think it's really valuable and interesting, quite frankly, that you also used to work for OpenAI. What can you just in a few minutes here, can you give us sort of a summary of what is or is there at all a meaningful intersection between quantum and all the work on the AI front that's been going on?

Aaronson: Yeah. So, certainly there's an intersection. And, you know, people have been thinking about this for years now. The stuff that I did at OpenAI had nothing to do with quantum computing. Right. So, you know, AI is now, you know, changing civilization in ways that, you know, everyone can now see just purely using, you know, massive amounts of classical compute. Right.

And not only that, it's not, but it's not even obvious how much a quantum computer would really help for tasks like training a neural net or, things like, you know, like probably, you know, because of Grover's algorithm and other things, there would be some advantages. But, you know, we don't really know if huge ones like the one from Shor’s algorithm. Huh?

Farshchi: It would at least help on the inference side, one would assume, right?

Aaronson: It might, but the big problem is that for most AI problems, we only know how to get a Grover-type speedup. Right? And a Grover-type speedup, eventually it matters. But, you know, because of the enormous overheads from doing quantum error correction, it'll probably be quite a long time before that becomes a win in practice.

Farshchi: Yeah, that’s fair.

Aaronson: Okay, so, you know, so I think, you know, finding killer apps where quantum computers can give you exponential speedups for AI problems that actually matter in practice, you know, that remains like an active research area, right? People, you know, are looking, you know, including my students, you know, my group. You know, we look for such applications. But I don't want to misrepresent the state of the field to the public and tell them that we have that killer app when the truth is that we don't yet. Okay.

But now in the other direction, you know, AI is also, you know, helping with quantum computing. Okay. So, you know, to do the quantum error correction; to detect which qubits have failed. What do I have to do to fix them, right? This is like: this places enormous, demands on classical computing, right. We're going to need ultra-fast classical computers to be constantly monitoring the qubits and seeing which ones have failed with, you know, how do we correct the errors? And, you know, very recent work has shown that you can get big improvements on that kind of task using neural nets, using deep learning.

So I think, you know, AI is going to help with the problems of building quantum computers, as it's going to help with just about everything else that humans do. It's, you know, it's hard to put any limit there.

Vorndran: So we are going to have to stop there. Scott, I can't thank you enough. I feel like I'm back in my engineering program in college for the first time in close to 30 years, which is did allow me to dust off the rust of my brain about some things I knew about that I probably forgot, and I learned a whole lot about a lot of things I didn't know, but just want to sincerely thank you for your time and quite frankly, your research that's, so important for the generation and the evolution of our country.

Jamil, any closing thoughts?

Farshchi: Yeah. Scott, this has been fantastic. I learned a lot. Thank you for everything that you're doing. I got to say, I've been, I picked up the show on Apple TV. It's called the “Foundation.” I watched it the other day. There's only a couple of episodes on it, and it talks about.

Aaronson: The Isaac Asimov series?

Farshchi: No. Oh…

Aaronson: Oh. Different one. Okay, I don't know. Okay.

Farshchi: Anyway, it's about this mathematician who is exploring factoring, and then everyone in the world tries to come and kill him. And so all I would say to you is watch your back, brother. You're doing some great work.

Aaronson: Okay. All right. I'll try to watch my back, but, you know, but speaking of which, thank you for everything that you do at the FBI to keep this country safe. And thank you for having me. It was an honor to be here.

Vorndran: Thanks, Scott.

Farshchi: Appreciate it, man.

Aaronson: Yeah. All right. Great.

***

Vorndran: Well, Jamil, I think I need to go take a mental break after that conversation. A lot of deep knowledge there on quantum technical language. But I think, you know, for me, the takeaways are really straightforward. I don't know that I can boil it down to three, though, right?

Perhaps number one is the lack of commitment to a time frame for quantum, right? I think that we always—I get asked this when I teach, right, what is the timeline? Here's what I saw in the media about Google, etc. And Scott was pretty specific that there is no dedicated timeline right now. There is no predictable timeline because of the variability of quantum mechanics.

A second one for me, and I'll probably go three deep here, and then go to you. The second one is this continued concern about, while only, you know, as Scott mentioned, Diffie-Hellman, elliptic curve and RSA encryption can likely be broken by functional quantum. Right. It's what have the Chinese what have the Russians will have the Iranians or other adversaries stolen from the United States, whether it's from the US government that's classified or not classified, but also from the American public, that they will be able to decrypt and potentially leverage, against our country?

And for me, that is a real, real concern. Certainly when we look at the Chinese hacks of Anthem and others, O.P. the OPM hack and the amount and mass amount of data they have that that remains are really, really concerning for me.

And then the third one, which I already touched on, is there is this broad belief that quantum will allow for the decryption, the refactoring of all encryption, right? And Scott was pretty specific that that's not necessarily true, right? That really what we're focusing on here that is at risk is Diffie-Hellman, elliptic curve, and RSA encryption standards. Which gives me a sense of comfort.

And, you know, as we close out this episode, after I go to Jamil, we will hang the quantum NIST standards, the post-quantum encryption standards, as part of this YouTube session so that people have access to it.

But, Jamil, your thoughts?

Farshchi: I don't know that the symmetric versus asymmetric gave me a ton of additional—I mean, I think it's nice that it's sort of it's half of the equation, but the problem is the half of the equation that this affects runs the internet. It's I mean, this is…

Vorndran: Sure.

Farshchi: And it's kind of hard to do stuff with purely symmetric. Plus symmetric comes with its own risks, as well. What I think what would encourage me a little bit about this is the…is the timeline. Like, I mean, he didn't seem as if this is around the corner at all. And, you know, it feels like every year that goes by, there's some prognosticator that suggests it's right around the corner. It's just a few years out or whatever. He suggested that it might take quite a bit longer than maybe what a lot of folks are anticipating. So that's encouraging.

And the reason it's encouraging is because there's just such a ton of work for us to do on the implementation side, to be able to roll out these quantum-proof mechanisms and the infrastructure necessary to be able to secure ourselves in advance of it.

I think the past is the past, and so the stuff that you were referencing, that the bad actors and the other nation states have already collected, you know, that's water under the bridge. There’s not much we can do about that at this point. But we can stop and we can advance and protect ourselves for everything going forward. So I think we need to put a lot more effort behind ensuring that we are protecting ourselves on this front and making the right investments and prioritizing it to be able to get ahead of this. Because it is inevitable that it's going to occur.

I think the last one I would say is that, this was interesting. I feel like quantum has been discussed forever. AI is more relatively new, at least in its current state. And, I thought it was interesting how he highlighted the fact that AI itself actually will serve as an enabler and an accelerator for the development and, and, productionalization of quantum itself.

And so, you know, that that in and of itself may shrink that time frame, which puts more pressure on us to drive my…toward my second point around, the investments that we need there. But, I guess we'll see. We just need to make sure that it's continues to be a priority for all of us.

My last point is, look, I'm glad we've got people like Scott on our team that are doing the work that they do. Super smart folks who are on the cutting edge of theory to be able to try to solve for this stuff and put us in the best position possible. It's…super smart guy, really interesting dialogue. And I'm looking forward to seeing what he and his colleagues continue to advance on in the future.

Vorndran: Well, thanks, Jamil. A public thank you to you as a Sooner for getting along with a Longhorn professionally for the last hour. For our audience: Thanks for joining us on Ahead of the Threat. And thanks for helping us get ahead of the threat. We'll see you next time.

Farshchi: Thank you.