Cloudflare CEO Matthew Prince: AI found a critical PHP vulnerability hitting 70% of the web — and millions of attacks are happening per minute

Speaker 1: two shell. I'm I'm rusty. I wasn't here yesterday. I'm back today, but we're very excited to be joined by Matthew Prince. Let's bring him Sick day. Whatever that means. Here he is. Yes. I was out yesterday. I'm a little rusty. Thank you so much for taking the time, Matthew. How are you doing? I'm I'm good. I'm... I mean, it's been a it's been a busy Yeah. Been a busy week and and and especially busy morning getting getting the news I'm sure. Of this out, but really proud of the team at Cloudflare and what we're doing to help help make sure that the Internet stays safe. Yeah. What actually... What what is the timeline here? What happened? When did this start in your world? Because I think a lot of people found out about it from your post yesterday or from, a number of other posts that went out just yesterday, but, it seems like your team is not asleep at the wheel. You've known that this is gonna be the year of cybersecurity risk. You've been doing a lot of things. So take us through your timeline of what Wearing my tinfoil hat. You guys were having some big party yesterday. We were having we were having... We had our we had our customer conference as well. So so it was it was... Customer conference. All I saw Sounds like a victory last of me.

Speaker 3: You go. We we had a we had a good time. Yeah. But... So so we were about two months ago Okay. OpenAI originally reached out to us to let us know that they, in combination using their models but also some of the work that some security researchers had done Yeah. Had found a vulnerability in a piece of software called PHP. PHP is a programming language for the web. It's a very popular programming language. About 70% of the web uses it. Popular CMS services like WordPress are built on top of PHP. And the problem with this vulnerability is it was called an RCE or remote code execution vulnerability. And what that means is that you can send some basic just requests to a server and then essentially take that server over, be able to run your own code on that server. So you could read all the files on the server. You could spy on anything. If someone was typing in usernames and passwords on a website and you'd taken over the server, you'd be able to see those usernames and passwords. You could take it over to do other criminal things like make it part of a botnet, launch denial of service attacks, distribute child pornography. And so the ability to take over a server is extremely bad. And this vulnerability allowed that to happen. Not that long afterwards, Wizz, which is a subsidiary of Google, also discovered a second vulnerability in PHP, which they also disclosed to us. And I think in this case, Google's team and OpenAI's team did exactly the right thing. They quietly worked with industry leaders like Cloudflare to say, hey, we found this problem. How can we clean it up? And it wasn't just They went to Automatic which is the creator of WordPress. They went to the PHP Foundation which maintains PHP. They reported it to the United States government. And over the last two months, there's been a concerted effort by lots of organizations to say how can we get in front of this particular risk. And so at Cloudflare, we implemented for all of our customers for free, quietly behind the scenes without really letting anyone know a level of protection. And that was true for even our free customers because we think it's important to make sure that the internet stays safe no matter what that help protect against that. And that's effectively like a band aid. You know, if you break a bone, you want to actually get the bone fixed. But in the meantime, you might put a sort of a bandage on it. Clefler provided that bandage to help buy some time. But anyone that's running a web server today, you know, the the... You you need to be up... If it's running PHP, you need to be up... Updating that. It's it's critically important. The NSA has issued guidance that if for some reason you can't issue the patch, it's actually better to take your server offline than it is to to run it naked right now. And we we went from seeing, you know, before, you know, or... It's clear that some details of the of the vulnerability got leaked, which if you're reporting to that many people, fact that that was kept secret for as long as it was. Yeah. I was gonna say, that is that somewhat...

Speaker 2: Yeah. Isn't that somewhat of a miracle that this has been going on for two months and it required coordination between people and tons of different country... Countries and different organizations. Like, honestly, some... You know, restore some faith

Speaker 3: in humanity that people could keep that big of a secret that And it was... And I I think they did. It was it was just in the last, you know, thirty six hours that we started to see an uptick where clearly something had leaked. Interesting. And and that then prompted the PHP Foundation to release the patch. The problem is the minute you release the patch, it gives enough detail that... And people can figure out how the exploit happens. So we went from hours before the patch was released, we were seeing thousands of attacks across our network. At this point, we're now seeing millions of attacks per minute Wow. Across our network where there... You know, people are trying to compromise servers. And again, you know, Cloudflare's protection is helping. But the real key is if you're running a server right now, get it updated with with the latest version of PHP. It's free. It should only take a few minutes to do it. Sure. If you're if you're using managed software like like WordPress, often it will do it for you automatically.

Speaker 1: But that's that's the most important thing that you can do to stay safe right now. What about in the in the consumer realm? Are there... I've I've been sort of... You know, I've I've run into other people who have been like, oh, I've heard about this. Do I need to update my iPhone or update my Chrome tabs? Like, is there anything that that that breaks out into the consumer world, or is all of this just for, webmasters, people who are actually serving,

Speaker 3: web services, running servers, or does anything bleed over into consumer just yet? People are saying cycle your your Social Security number. Oh, yeah. Rotate your Social Security number. I saw that. Yeah. I think... I mean, I I think that in this particular case, you know, the PHP vulnerability Yeah. Really applies to businesses and people running running web servers. So there's not as much for the consumer. But but, you know, it would... As as if TSTO and the Chainsmokers and Common and a... And one of the worst vulnerabilities the web has ever seen wasn't enough for one week. We also had other vulnerabilities that were disclosed this week. There was a Node. Js vulnerability that was also something that probably consumers have to care about. The one that consumers should care about more is that there was a... It was discovered that one of the key root servers that powers the encryption behind the Internet, they're supposed to be secured with these very long complicated passphrases. They're supposed to be three twenty eight bits at a minimum. Instead, an an Indian company that was running some of these these root root server... Root key servers Yeah. They... Instead of using 328 bits, they used literally the number four. Yeah. So their password was less secure than your ATM PIN. Yeah. And the problem with that is that because... Can we give them the the benefit of some four d chess?

Speaker 2: Because... Who would think to guess that? No one... No one's thinking if you were gonna try to brute force it, you would pick a number at least in the four digits

Speaker 3: Yeah. And then go from there. They had they had, I I think... I I can't remember if six or eight of their... These these key... These root these key root server certificates. The the most secure password they used was was the number 136. So they did... You know, they they they they... Again, almost to your ATM PIN Yeah. At that point. They call it security through obscurity.

Speaker 2: No one would... And they just rage they just rage quit the product category. Did I see that? I think they just... Yeah. They did. So... Which I mean... You can't fire us. You can't fire us. We quit. How

Speaker 3: you can ever be trusted is is is tough. But this is actually one where consumers do have something because Sure. These certificates, if somebody could get in the traffic flow, they could pretend to be Google, they could pretend to be your bank, they could trick you into doing a bunch of things. And so what the industry has done is as soon as this happened, your browser, your operating system distrusted these particular certificates. Sure. But until you upgrade your browser, until you upgrade your operating system, then you might still be trusting these certificates and your and your browser could be spoofed. So I think that that right now is a great time when your when your phone says, hey, do you want me to upgrade overnight? The answer is yes. Yeah. When your browser is harassing you to, like, you know, upgrade, like, just do it. Takes a few seconds. You're gonna be much more secure. And I wish I could say that like this was an unusual week and we're not we're never going see another week like it. You're normal. I think we're going see a lot more weeks like this. I think the next two years are going to be incredibly, incredibly, incredibly fraught with what again are these new tools that we have that have infinite patience to go look for vulnerabilities, finding real problems across the internet. And it's going to feel scary. It's going to feel like the sky is falling. It's going to feel like there's nothing we can do. And I think there are gonna be stories that characterize AI as the bad guy. AI is not the bad guy here. AI is the good guy. The AI didn't create these vulnerabilities. AI found them. I think the question is like, if you're living in a house that's got deadly monsters living in every closet, but you don't know they're there, are you are you in a in a better place or a worse place than if you start like going through the closets and finding the monsters and killing them? Yeah. Like, it's gonna be a lot more frightening to go through and kill the monsters, but you're probably gonna get a lot safer. And so I think we're gonna see some very scary stories. But there you go. Scary monster music. We're gonna see some very scary stories. But the good news is that, you know, you're... OpenAI, Anthropic, the Frontier Labs, and then in partnership with companies like Cloudflare are actually killing the monsters off. And I think the Internet is gonna get more secure. How are how are attacks evolving?

Speaker 2: Like, are are you seeing, like, swarms in in a way... In in, like, in a way that's, like... Swarm is obviously, like, this sort of, like, it's a it's a a real thing, but it's also kind of an idea. It sounds like something out of The Matrix. Yeah, like something... Is it that different than historical efforts when there's this kind of vulnerability? Because I imagine you can just get effectively a directory of every possible website that might have this vulnerability

Speaker 1: and then run a program. Yeah. There's something called a rainbow table. It's basically like every possible password. And every possible password. Yeah. But that was easily mitigated. But, yeah, a for loop with every possible... Are seeing like

Speaker 2: rogue swarms of of... Like powered by open weight models that are that are driving a lot of this? Or is it just, hey, there there's vulnerability

Speaker 3: that impacts 70% of the Internet? Mhmm. Yeah. I mean... So so first of all, you don't need AI to take advantage of the vulnerability in this case once it's been discovered. So the the people who are trying to exploit this now, they're not using models to exploit it. They're just running they're running, you know, fairly simple scripts that are just, you know, trying to check every single server, every single IP address online Yep. To see if it's vulnerable to take it over. Mhmm. What what I think has changed is the rate at which attackers can figure out what they do when they find themselves in an environment. So about a year ago, there was a very, very, very popular plug in that plugged into Salesforce that got compromised by a Russian hacker. And it was it was so broadly deployed that it effectively gave the Russian hacker access to a huge portion of Salesforce's infrastructure and Salesforce's customers. But what was interesting was when you actually studied what the attacker did, they clearly... Like, they they weren't in enterprise sales. So they didn't really know how Salesforce worked. Sure. And so they kind of bumbled around like not quite knowing what to do or the fact that they'd stumbled on what was an incredible treasure mine. And as a result, what could have been just a disaster of a hack wasn't nearly as bad. What we're seeing now is if that same hack happened today, the attackers are using AI to say, okay, I found myself in Salesforce. What what stuff should I go, you know, looking for or stealing or trying to do? So the ability to very quickly, you know, figure out I'm here. What what what can I take advantage of, You know, is gotten there? What we're not seeing yet are, you know, I think that the example out of OpenAI with the hugging face compromise, I mean, that again makes, I mean, it's almost out of science fiction, some of the things that are happening there. But I think the good news is that the cost of running that type of Intact, an agent swarm like that is still extremely expensive. Hundreds of thousands, if not millions of dollars for that. And so an organization like, you know, OpenAI that's got those resources, the good guys have more resources than the bad guys. Yep. And so for now, I think you're you're actually seeing kind of a very naive use of AI by by at least the sort of typical cyber criminal. He's probably different when you're looking at nation state actors. I think that if you if you look at China or Iran, the NSA in in The United States, like, that... That's a different case. But but right now, I think that AI is actually helping the good guys more than it's helping the bad guys. Yeah. We're finding the monsters that are hiding in the closets and... What advice

Speaker 2: would you... So over the last week, we had open... Like, a open code ran into issues and reported being, like, attacked for basically days on end. Obviously, the attackers were going after the... It was token theft. Yeah. Effectively token theft. It's the modern version of... Railway. Railways had had massive issues over the last twenty four hours. I know of another big inference provider that has been, like, under attack and, like, viscerally feeling that for the last three, four days. Wow. And that that that people... That everyone And would

Speaker 3: this is... And what this is probably... What a lot of these are symptoms of are the attackers are like, wow. We could do a lot more if we had more compute, if we had more AI models. And so stealing tokens Yeah. You know, hacking inference providers, That... That's that's giving them access to tools. We... You know, you'd see this before. Like, it's... For a long time, Cloudflare was an Okta customer, and Okta kept getting hacked over and over again. And the reason why... And Okta was an identity provider. I mean, the the reason why was they were going after us and they were going after Coinbase, which is another another one of their one of their customers. And again, I think you... The attackers are trying to find their way to whatever is valuable that they can that they can get to online. And and I think right now, the ability to have more inference, more tokens is something that attackers are very much longing for. And so they're going to go after a bunch of those providers. And then to your question of what to do, like you got to just stay up on all of the security patches that are out there. You know, my... I I don't know the the the details of, like, the railway attack, but my hunch is it's probably based on one of these, you know, latest vulnerabilities that was out there. And so it's just, you you know, they... You talk about like an ODA loop, which is like how quickly you can adapt versus your attacker. And so at Cleflo, we're just sort of saying how can we make that ODA loop faster and faster and faster and faster so that we can adjust incredibly quickly. And part of that is using AI tools and using things that can actually proactively you know, have a swarm of defenders that are looking out for the swarm of attackers. And and that's that's exactly what we're doing at at Cloudflare and what we're doing for all of our customers. Do you think that ski resorts are ready for this new

Speaker 2: wave of cyber vulnerabilities. I'm I'm joking, but I'm joking, but I'm actually... Yeah. People have been talking about bringing down the grid. If you bring down a ski resort, that's actually... You're out Very popular. You're out getting a couple laps in and the I mean, I was... And the lift signs to a halt, your your first thought might be, oh, is this kind of like a ransom attack? Or the

Speaker 3: the snowmaker turns on when it's not below freezing and just sprays water all over everyone. It could be a horrible thing. I mean, and I think that, you know, I I I have some insight into into a into a certain ski resort operator called Vail and and some of the the board meeting conversations with the activists. And and one one of them, the... An alternate person for for this lady, named Bryce Roberts, who's a great VC. I think you've had him on. Yeah. We love Bryce. He he he... He's he's up to to to to be a candidate unveiling. And he said, have you guys thought about AI at all, like, in terms of that? Because it's something like, you know, 30% of questions to AI are about trip planning. And like literally everyone in the room scratched their heads and said, oh, we haven't thought about that at all. There's a there's a lot of there's a lot of there's a lot of industries that that need some serious modernizing and and a lot of management teams that need some serious modernizing. And I might point at Veil as being one of those. Yeah. Last question for me. It seems like the balance between attackers and defenders is going to be increasingly

Speaker 1: infrastructure driven, and I feel like this might change the discussion a little bit, maybe not at a national level, just around the AI data center build out because there's a certain amount of just if you have a bigger data center with more GPUs running more inference to secure systems and you're going up against a smaller data center with a similar model or a distilled model that's three months behind. Like, you want as many advantages as possible. And I'm wondering if you think that, like, raw compute power becomes an important discussion in, like, the geopolitics of cybersecurity.

Speaker 3: Yeah. You know, I mean, maybe at some level. I think that, again, I think I'm encouraged that it's the good guys have more resources than the bad guys. And that's, I think, that's going to help the defenders stay in touch. But you've got to have great partnerships between companies like OpenAI and Anthropic, companies like Cloudflare and others in order to make sure that this is going forward. At some level, the whole thesis of Cloudflare from the beginning was that if you got enough internet traffic flowing through a system and then you could use machine learning to look for patterns and attacks and threats that you could that you could see things and stop things. And I think we're really good whether that's an attacker working for North Korea or it's a swarm of agents at identifying those things and staying in touch with it. And I think that's why we're seeing such just real demand for our services and more and more people signing up for Cloudflare. And again, I think it's important to us to say that you don't have to pay us in order to stay safe online. We want to make sure that everyone has the resources regardless of what their budget is. And so we continue to be committed to this. And again, I wish I could say that this was like a really unusual or bad week, but I think for the next two years, we're going to basically have kind of a week like this on average almost every week where big vulnerabilities are getting released. We already know about a number of substantial other vulnerabilities that will come out over the, you know, coming weeks or months. Sure. And and the good news is that we're ahead of them and we're and we're protecting them. And if you wanna make sure that you're protected,

Speaker 2: using a service like Cloudflare is one of the best kind of, you know, bandwidth. Yeah. Not... You're not making a prediction. It's like you see the pipeline, so you know this is gonna happen. And then it also, like, the more the more high profile this trend gets, the more motivated people are gonna be to find the next one. At least one of the security researchers that at least claims to be involved said that it was found originally by type prompting find bad vulnerabilities in PHP. No way. That really is? Someone's claiming that who

Speaker 3: claims to be involved. It's crazy. Who knows? But... Yes. I mean, but it's it's... Again, like... And it's it's it's gonna feel really scary. We're gonna find a lot of monsters and a lot of closets. But the monsters were there already. Yeah. Like, they were they were they were just waiting to come out and kill us when we weren't expecting it. I would rather we actually be turning it over and and do it. So I think... Do you think there there has to be there has to be some groups that are really upset because they had found these vulnerabilities and have been exploiting them? People are talking about that. They're like they didn't want... No. They didn't want it to get out. It's gotta be... They're they're... I mean, I talked to folks at the NSA that are like, this sucks. Like, you guys are finding... I was using this to spy on nation's enemies. Yeah. And so... And that's and that's... Again, I think that that's a... I think that's a... That's... That is all... Consequence of modernity. Yeah. That's all good news. Yeah. Because because we're finding the monsters. We're killing the monsters. It's gonna feel very, very scary. Yeah. It's gonna feel like the sky is falling. But I think, you know, we're gonna come out of this. Yeah. And two years from now, I think it's going to be almost impossible to write buggy code. At least in terms of the sort of classic divide by zero or off by one or unconstrained memory glitches. There'll still be business logic code that that people will get wrong or or machines will get wrong. But, like, already, we see how many of the errors AI is helping us correct. Mhmm. And so even though I think we're gonna have a a a couple of years of really, really, really bad stories. Yeah. I think that the real trend, the real story that's not gonna be popular to say but is true is the Internet's gonna get a lot more secure and we're all gonna be better off because of these tools and how they're helping protect. That's amazing. That's amazing. I have one last question. More years of the most bullish headlines for Cloudflare.

Speaker 1: Gonna be There's... There there there was a lot of debate throughout the year over the the the pros and cons of these staged rollouts. Like, Mythos goes super viral, definitely draws people's attention, but a lot of people are like, ah, I I wanna use that for my business. And then Fable comes out, and it's sort of nerfed in cybersecurity, and then Daybreak comes out. And there's there's different ways... Glasswing. There's different ways to get into these programs. Looking backwards, it feels like I like the idea of, you know, everyone gets amazing AI, but I also like the idea of everything getting patched before the before the negative things affect me. I don't necessarily wanna need to, you know, inference the most advanced model just to, like, use the web myself and write... Hack into the service that I'm using to patch it before I give it my Social Security number, etcetera. Like, are you happy with the current flow of intelligence breakthroughs staging out, getting to you a couple of weeks early before something else happens? Is the current model sustainable? Is it correct? Or does there need to be a rejiggering there? Yes. I would say that

Speaker 3: you know, from our perspective at Cloudflare, we have loved the partnerships that we've had with OpenAI and Thropic, the other I leading frontier think we've gotten trusted access to a lot of the models ahead of time. And then we've been able to use that to help protect more of the web and work in coordination with that. But I do think that it's a tough problem. Like when I meet with European banks or leaders and they're like, well, why don't we get any of these things? I mean, they feel very vulnerable it's a tough thing to figure out exactly what you do. In this particular case, there are a lot of people who are like, well, why didn't you tell me about this PHP bug And it's hard to get that perfectly right. The more people you tell, the more access that you give, the more risk there is that it falls into the hands of someone who will do real harm with it. And so, it's a game of trust in a lot of ways. And I remember when Clavicle was a small startup, you know, we were always like, why aren't you telling us about the new security vulnerabilities? Trust us, trust us, trust us. And now we're a much bigger company and we do get those things and we get and I get calls from startups like, well, how can I get it too? Sure. And I think it's tough. Yeah. I I don't know that there's a perfect answer. I I I will say that the fact that this, you know, was disclosed almost two months ago that it gave a lot of of organizations the ability to get in front of it and patch it. That's gonna have mitigated what could have been just an unmitigated disaster. And again, hats off to the team at OpenAI, the team at Google and Wiz for how they handled this. And I think there will always be people who will say you could have done something a little bit better. But I think this is... We're not gonna be perfect and this was pretty darn good. Well... What is your third final...

Speaker 2: Well, what is your answer to our third final question? What is your most A mutual friend of ours texted me a minute ago asking me to ask you what is your most outlandish prediction for the Internet in a decade?

Speaker 3: The business model is going to change fundamentally. So we're we're we've got we're maybe this about to have this massive tragedy at a commons problem. So if current trends continue in five years, automated traffic, agent traffic is going to be a thousand times human traffic. And to give you some sense, the internet almost collapsed when during COVID traffic doubled. So a thousand X going to be, you know, somewhat, there's a lot more servers, there's a lot more routers, there's a lot more bandwidth that has to get provisioned in order for that, to work. And the question is who's going to pay for it? Because if I go to my chat GPT agent right now and say, hey, where should I have lunch? It might check a thousand different menus across a thousand different restaurants. And that's burning resources from all of those different places. And yet it only is going to send me to one of them. And so my hunch is that something's got to change in terms of how the internet works. And my hunch is probably some version of micro payments that every time you access a resource, your agent pays a 10 thousandth of a penny to be able to see the menu. And then there's probably some sort of a trade where if you are willing to tell the restaurant what it is that you're looking for, what what your agent is doing, you're willing to reveal something, then that price goes down maybe even is free. In some cases, the the site might actually pay you to come to come visit it. Yeah. But that that will then lead to this kind of gonna be the next version of advertising, which is how do you advertise to agents? If I know that an agent is coming to my travel site and they're specifically, you know, on behalf of their user looking for a Hawaiian vacation to a five star resort. And I know that there's a five star resort on Maui that has availability and a good price, like matching that together, I think is going to be a resource. That's going to be, I think, big part of how we've solved what is otherwise just in a massive tragedy, the commons problem. Because otherwise, like why wouldn't your agent check literally every website on earth to come back and say, you know what? For lunch today, you should go to Wendy's. Yeah. So and that's like, that's not that's not a good use of time. So I think that's gonna be the the the most interesting thing to figure out. And I think the most interesting question, certainly, the next five years, I think we'll probably have to figure it out before the end of the decade Yeah. Is is what's gonna be the future business model of the Internet. And and that's what we're spending a lot of time with. That's exciting. I love that. That was very exciting. I had a funny I had a funny moment the other day. I was doing some research on

Speaker 2: on this car, and I went to check the source. Like, I was doing it in Chateaune. I went to check the source, and it was on, like, recipes three six five. And clearly, someone had used AI to just, like, generate content for, like, a specific And I'm realizing, I I can't I can't lean on this for car

Speaker 3: At least you're looking at the sources. I mean, it's... We... We're seeing that fewer and fewer people are going back to the original sources. And and like for for like journalism, like journalism matters. Like what you guys do matters. And someone's gotta pay your bills. And in your case, like, it's OpenAI. And... But I think that's probably gonna be the case in a lot of cases. Like, how you still... You're still gonna need reporters that are going out Sense there and finding. Paying people and feeding it. And we've gotta have some system

Speaker 1: that that allows the the people who are doing that work to get compensated because I I think you guys deserve to eat. Thank you. Well, thank you so much for coming on. Great to see you. Great to hang out. Good luck out there. We'll talk to them again. We'll see you soon. Goodbye. Let me tell you about MongoDB.