a16z Podcast
Original podcast
Where to listenCopy share link
4,161 words · 3 speakers
Okay, cool. So let's get started. So welcome to the security session. Thanks everybody for coming early in the morning. This should be a fun session. So we have four speakers lined up. I guess I'm the moderator, but I'm also going to give the first talk. I'll talk a little bit about Bitcoin and some other work we've been doing. Then we have Kevin Fu is going to talk about security and healthcare devices. So if anyone here is wearing a pacemaker, you better be you'll be quite nervous after you hear Kevin's talk. Then we're gonna hear from Giovanni, who's gonna talk about malware analysis and web security in general, yes. And uh then Raluca will talk about privacy and um enforcing policy in the cloud. So it should be a fun session. We're gonna each talk for 15 minutes, so we should be done by 10, 1005 or so, and then we'll have time for 15 minutes QA, then we'll have a break, and then we'll run into the next session.
Um which is uh whatever it is. It's uh industry conversations. Yes, so that'll start at 10 30. All right. So with that, uh let me get started. So uh all four of us have actually done companies. Uh I'm actually a happy camper. My my company was uh acquired this year, so things are good. Um so instead of telling you about companies we've done, I wanted to tell you about kind of current research that we're doing. In particular, I'll tell you about two projects that uh we've done um just very very recently. So the first one is on Bitcoin, and you know when um when you go pitch to VCs, the first thing they they tell you is like on slide number one, you're supposed to have some word somewhere it's where you're supposed to have the words billions of dollars somewhere on slide number one. Otherwise the VCs don't pay attention to you. It's like the golden rule.
Yes? So fine. So my billions of dollar slides is gonna be on the Bitcoin capitalization. So you all know you've all heard Bitcoin. This is kind of a graph. You can see Bitcoin kind of launched in 2009. This is the amount of dollars invested in Bitcoin. You can see it kind of took off seriously in uh 2013, it reached quite a peak, like mid 2013. Everybody thought, oh my god, this is gonna take over the world. Uh no more dollars, everybody's gonna be doing bitcoins. Things kind of changed. It was a bit a bit of a rough time in the last uh year and a half. So that's the beauty of technology. You can't ever quite anticipate how things are gonna go. And today, the it's you can see that in the last six months or so things have kind of stabilized, and um market capitalization is around uh three and a half billion. Still very, very exciting. This is the first cryptocurrency currency that's actually seems to work that's actually been being adopted.
We've had lots of cryptocurrencies in the past, but none of them have worked. Yeah, they've all kind of work experiments that never really took off. Bitcoin is one is among the first to really really take off. Uh and in fact, I I I really find Bitcoin exciting. There's really lots to do here. Um, there's lots of crypto applications, lots of financial applications. So, in fact, we just started a class on Bitcoin. You can see CS251, it just started on Monday on blockchain technologies. Um the thing that we're excited about with bit about Bitcoin is A, the obviously the currency, but what's more interesting is even the block applications of the blockchain, right? It's the first time when we have sort of an append only database that's kind of um, you can put things on it, it's publicly available, and no one can take uh uh elements off of the blockchain, and that by itself has enormous number of applications.
So we're kind of excited about the blockchain and how to use it. So the course is gonna kind of is focusing kind of obviously on the beginning of Bitcoin and also on blockchains. Um there are actually startups who are helping us build uh labs based on in particular a startup called 21, funded by Andreasen Horowitz, um, that actually helped us build uh in uh a beautiful series of lab lab lectures and lab projects based on um Bitcoin and Bitcoin mining mining. So uh if you want to teach a Bitcoin class yourself, please use this as a resource. There's there's a lot of work that went into building the homeworks and the projects for this course. So I would encourage you to teach Bitcoin courses yourself at your own universities, and please you're you're very welcome to steal all the materials and and use it in your in your own classes.
Uh great. So unfortunately, there are bit you know, Bitcoin is not all rosy. There are some issues also with Bitcoin. Of course, you may have heard there's a certain uh Bitcoin exchange that asked people to give them their bitcoins and it just forgot to give the bitcoins back. Yes? So uh that was kind of an issue, right? That's kind of uh a bit of a disrupt disruption to this rosy world of Bitcoin. The question was basically so after this happened, the the question we were kind of curious about is well, is there a way in which an exchange can prove that it's solvent? In other words, the number of assets, the amount of assets that the exchange has, could it be more, could it prove that it has more assets than it has obligations? So this is the problem that we uh set out uh to solve, and in fact, we weren't the only ones looking at this.
Uh Greg Greg Maxwell also looked at this problem. So we were kind of wondering is there a way to ensure solvency? And essentially the problem is again, if you look at um what is the solvency problem, you have a Bitcoin exchange, it has a set of obligations, right? So there are customers who gave the exchange who gave the exchange their bitcoins, it's keeping the bitcoins for them. Uh so it has obligations to those customers, and then the exchange also has a number of assets, right? It all it holds its assets also in bitcoins. And solvency is basically the problem of proving that your assets are greater than your obligations. Yeah, that's solvency. And by the way, you can generalize it beyond solvency to actually talking about reserve. If you just put an eps uh an epsilon instead of in front of the obligations, you can, for example, argue, prove that the assets are more than um sorry, that uh what is it, that half the assets are more than the obligations, right?
So you can prove that you have a reserve instead of uh just a greater than or equal to. But I'll here I'll just focus on solvency, uh, not so much reverse reserve, it's actually the same problem. The issue is today, banks need to do this too, right? Banks also need to prove that they're solvent, and how do banks do this in the real world? Well, they hire an auditor, the auditor comes in, check the books, checks the books, and make sure the bank is solvent, and then the auditor is kind of trusted to keep everything secret and to be honest. Yes? So there's a lot of trust in this system, and there's also a lot of expense in that the banks have to pay the auditors to do this job. The beauty of Bitcoin is that you need no trust in third parties, and this proof of solvency can be done without uh any expense, any payments done to a third party.
Okay, so what our goal, what our goal here is basically basically to prove that the assets are greater than the obligations, but do it in a way that doesn't reveal anything about the inner workings of the exchange. So this is what's called a zero knowledge proof, right? I'd like to prove to you that I'm solvent, and I'd like to do it in a zero-knowledge manner. So you will learn nothing about my business, but you do, but you are convinced that I'm solvent. Okay, so that's what we uh set out to do, and this is joint work uh with uh two students, Deger and Buns and uh with Joe Bonot, Jeremy Clark, and uh myself. And basically it's an efficient zero knowledge proof for this problem. And like I said, if Mount Gox had run this protocol, their problems would have been detected much earlier than they then they were. So the idea is that a Bitcoin exchange would kind of run this every day, publish this proof, you know, delete all the old proofs, but every day they would publish a proof of solvency, and if they ever can't do the proof, that means there's a problem, right?
Then at least you know the public would know there's a problem. Maybe you should try to get your bitcoins out. All right, so I'm not gonna uh give you like all the details of how this proof works. That's uh probably is not appropriate for this audience, but I did go want to give you a sense of how the mechanism actually works. Like, how do you prove that you're solvent without revealing any anything else? So there are three steps to this to this process, and let me try and explain at a very, very high level how these uh how these steps work. So there are three sub protocols. So let's start with uh uh the the first two steps sub protocol one and two. So the mechanism that we use is kind of a standard mechanism that comes up in com in uh zero knowledge proofs, it's called a commitment. So a commitment, the way to think about this is I have a number X that I'm thinking about.
I can take my number X, put it in an envelope, seal the envelope, and give it to you. Yes? Now, obviously I'm committed to the number X because you know the X is in the envelope I just gave you, so I chan I can't change the number X anymore, but you have no idea what the number is, right? It's in a sealed envelope, you cannot open the envelope, so you don't know what number I committed to. Okay, so a commitment allows me to commit to a value, so I'm committed to it, but you don't know what value I just committed to. Okay, so commitments are a useful uh concept in general. Here we're gonna use them to prove uh solvency. Okay, so the first thing that the exchange is gonna do is it's gonna create commitment, a commitment to all of its obligations. Okay, so this means I'm gonna sum up over all the balances of all my customers, and I'm gonna commit to that sum.
Okay, so now I've I've told you basically what my total obligations are. Uh as we said, the commitment is by binding, but it reveals nothing about how many customers I have, what's the total number of assets that uh that that I owe to these customers and so on. Now, the beauty of these commitments is every user can actually be given some sort of a secret value. So the user logs into his account at the exchange, the exchange gives him some sort of a secret key, and using this secret key, he can actually verify for himself that his balance was included in the commitment to all the obligations. Okay, now how we do this, I'm not gonna explain to you, but just believe me that we can make sure we can make it so that every customer can be convinced that his balance is included in the total set of obligations. So if every customer checks that his balance is included, then we know that the set of obligations is in fact what it should be.
But in fact, not every customer needs to check. It's enough that say one percent of the customers check. If the exchange cheats and actually does not include people's obligations, you know, if a small number of fraction of the uh uh customers actually check that the proof is valid, uh uh a corrupt exchange will be quickly caught because uh a customer will see that his balance is not included, and then he can complain, and he has actually has a proof that the exchange cheated him. Okay, so that's step number one, committing to the obligations. Step number two is committing to the total assets. So now, now here's a this this is kind of a difficult problem, right? So now the exchange has to prove how much money it actually has. But again, it's gonna do it using these commitments. It's not gonna tell you how many bitcoins it has. It's just gonna give you a commitment to the number of bitcoins that it has.
So how do you prove, yeah, this is kind of bizarre, right? How do you prove um how many bitcoins I have without actually telling you which bitcoins are mine and how many they are there are uh total. So the way we do it is as follows. And this is something that's unique to Bitcoin. This is I think kind of the the the tricky part of this of this protocol. Something that you cannot do in the real world, but you can do with Bitcoin. So remember how Bitcoin works, right? In Bitcoin, there's this public blockchain. Every uh address in Bitcoin essentially is a public key for a signature system. Yeah, so there are all these public keys out there. The exchange owns a subset of those public keys, right? Those are the bitcoins that the exchange actually owns. Uh, and every all the other public keys belong to other Bitcoin users. Okay. Now, what we'd like to do basically is is here.
So, what the exchange does actually is it's gonna choose um, it's just gonna go to the bit to the blockchain, it's gonna choose like a million public keys. So, n here could be a large number. It's gonna choose at like a million uh different public keys that have all sorts of balances associated with them. The exchange doesn't know this, doesn't own all those million public keys. Maybe it owns a thousand of them. Maybe it only owns 10,000 of them. Yes, it only owns a subset of those public keys. What it does is now it can prove, again, and using this fancy zero knowledge technology that it knows the secret key for a subset of those million public keys, and that the sum of the balances associated with those public keys is is the number is the assets that it just committed to. Okay, so I'll say it again, it's a little tricky. So we have a million public keys that were just pulled off the blockchain.
Yeah, so these are like random uh Bitcoin addresses. They belong to all sorts of random people in the universe. The exchange is gonna prove that it owns a thousand of those public keys, and the sum of the balances associated with those public keys is equal to the value to the commitment to the assets that it just gave you. So this doesn't reveal how many assets it has. It doesn't reveal which public keys the exchange own. It just proves that there is a subset for which it knows the secret keys, and the balances associated with that subset is actually the set of assets, is actually the set of assets that the exchange committed to. That's the idea. Yeah? So now how we do this, like I said, there's uh uh a lot of technical detail that I'm gonna skip over, but just trust me that we can do this. So I hope this is clear. So now we've qu we've created a commitment to the obligations.
We've created a commitment to the assets. The only thing we have to do now is prove that assets are bigger than commitments, sorry, assets are bigger than obligations, and it turns out that's actually quite straightforward. Yeah, this is like standard off-the-shelf technology. There's a way that if I can give I can give you two envelopes, so two commitments. So I'll give you a commitment to X, so X is an envelope and Y is in an envelope, and then I can prove to you that X is bigger than Y without revealing anything about X and Y. This is just crypto magic that's old, and that's that's that's what we would use here to prove that in fact the exchange is solvent. So the cute part of this uh of uh so step number three is not interesting. Step number two is kind of the cute part uh that makes this uh that makes this problem interesting.
Absolutely.
Um in that thousand, what if outside of that set the exchange at the end of its biggest client where they're holding most of the bitpoints isn't in that thousand. So they're only proving as sort of
a small subsection of that. Oh uh actually its customers are included in protocol one, right? Those are in the obligations. So the exchange basically collects bitcoins from everybody, yeah. So it has like that, let's say, you know, coin that uh Coinbase, they have like you know, half a million customers, say it collects uh well, actually, I don't know how many customers Coinbase, I'm just making that up. Uh so say it has half a million customers, it collects all the bitcoins from all its customers, and it stores those bitcoins in in in bitcoins that it owns, and in addresses that it owns. Yeah. Um what it's proving now is that the Bitcoin that the addresses that it owns, uh well, it's committing to the sum of the balances with the addresses that it owns. So it's not, I mean, you set this this protocol two is not related to what customers gave it.
This is just basically how the bitcoins are stored. Yeah? Cool. Excellent. All right. Very good. So uh fine. So that's at a very high level how this protocol works. Now we went ahead and actually built this, of course. And uh so let me just show you how this how this works. So here are three lines here. What the blue line shows you on the bottom, let's just look at say you have half a million customers. So let's just look at the um rightmost uh side of this graph. So you can see that verifying the proof is actually really fast. So on a single laptop, a customer can verify the entire proof in about 10 minutes. And in fact, a customer doesn't even have to verify the entire proof. Really, the customer is just gonna verify that it's uh its uh uh balance was included in the commitment to the obligations and that the assets are valid. Yeah, so a c customer is actually, if they wanted to verify everything, it would take 10 minutes, but in reality you only have to verify a small slice.
Uh very uh sorry, that was actually, I'm sorry, the blue line is the time to generate the proof. So bit so the exchange will take 10 questions. Just a second. So the exchange will take 10 minutes to generate the proof. Just 10 minutes. Yeah, question.
So uh if the proof fails, it just might mean that the sample that it shows wasn't large enough to prove that a set's a lot of
no, no, no, no, no, because remember, in the set of Bitcoin addresses that it that it used, it's gonna have all of its addresses in that set. So necessarily the set is large enough, necessarily. What what is not being revealed is which of which subset of addresses it owns. Yeah, so imagine you you could even in principle take all the addresses on the blockchain and prove, so that would be the the anonymity sets, yeah, and then prove that you own a subset of those and nothing else would be revealed about your your set. Yeah? So it's only zero now in the limit the size of payments. It's sufficiently big. Well, it doesn't have to be it depending. Now it's a policy matter, right? If you're if you're comfortable in saying that what that you know a few of these million addresses are mine, then that's fine. If you want to reveal nothing, you would use the entire blockchain.
And then you just say, you know, some of these addresses are mine, but that's already known. Yeah. Cool. That's a that's a really good point. Okay, so verifying the proof, let's see, that takes about 40 minutes, verifying the entire proof. That's the orange line, takes about 40 minutes, but again, you wouldn't have to verify the entire proof. You would just verify a slice of it. And then the proof itself is you know, for half a million customers, you know, it's about a gigabyte. So every every day the exchange generates a gigabyte, stores it somewhere, throws away yesterday's proof. So storing a gigabyte in the cloud is nothing. That's pretty easy. Every customer just downloads its slice of the of the proof, verifies it, and we're done. Okay, so that's uh that's how this uh scheme works. Uh so very practical. We've actually, well, we're trying now to get the the uh coin the Bitcoin exchanges to adopt this this style mechanism, and you know, we'll see, we'll see how it goes.
Okay, so I'm at the 15 minute mark, so I guess I'm gonna skip over the control flow integrity, unfortunately. This was kind of a fun project I was hoping to tell you about, but I'll skip over over that. I don't want to take too much time. So actually, I'll just tell you the punchline. Ah, come on. Why doesn't this go forward? I'll just tell you the punchline. The punchline of this part is AS is a block cipher. Intel did amazing service to humanity, and they implemented AS inside the X86. This was an inside in the their processor. This was an amazing success for Intel, to the point where now you see it every generation of microarchitecture, uh, they're actually improving it. So haswell is the old architecture. A yes, one round of AES took seven cycles. Skylake, which just came out this summer, now has been reduced to four cycles.
In two years, it's gonna get down to two cycles. So the message that I wanted to uh you to take away from this is two messages. First of all, block ciphers, we used to think of them as being expensive to evaluate. They're basically free because of what Intel did. This is kind of remarkable. And second, uh, it's gonna be really hard because of this, it's gonna be really hard to promote any other block cipher, because any other block cipher is gonna have to be implemented in software, and there's no way it will compete with what Intel did. So in some sense, AES is not gonna be with us for many, many years to come. Okay, so that's how I guess I'll stop here. Last thing I'll say is there are um ah, okay, I need 10 seconds. So last thing I'll say is um at Stanford, actually we have something something called the Cyber Initiative.
This is a big grant from the Hewlett Foundation that looks at the integration of of uh technology into society. We're running a whole bunch of events in this in this space. So if you'd like to join or if you'd like to speak, please let us know. So you can see the type of questions we're looking at are ethics of data collections, privacy versus security. Is there really a trade-off? This question of data ownership. So if you'd like to speak or if you'd like to just participate, you know, please come or please let me know, as I said, if you'd like to speak. Okay, with that, I will stop here, and I guess we'll move on to the next speaker. Why doesn't this advance? Here we go. We'll move on to the next speaker. So thank you very much. And
Page updated August 25, 2026 · transcript and summary are machine-generated