00:00:02 Speaker 1: Bloomberg Audio Studios. 00:00:04 Speaker 2: Podcasts. 00:00:05 Speaker 3: Radio. News. 00:00:09 Speaker 2: AI helped uncover a potentially catastrophic vulnerability hiding for years in PHP. It's a programming language used to build and run millions of websites and web applications. OpenAI's technology helped researchers find the flaw. Cloudflare helped protect websites before that flaw was publicly disclosed. But AI could also help hackers exploit it faster. The concern is that the flaws could potentially let a hacker steal data or seize control of websites. Joining us now, OpenAI President Greg Brockman and Cloudflare CEO Matthew Prince. I'm grateful to both of you for your time. Greg, I want to get specific here. So what happened was Calif used... GPT 5.6 Sol to find a version of this bug back in August. And then OpenAI independently kind of found a more complete version in September. What is it that OpenAI's models actually did or do that allowed you to find something that I think researchers have not found for like years and years? 00:01:10 Speaker 3: First of all, thank you for having me. You're exactly right. Calif had a security researcher use our models very guided by that researcher's expertise and intuition. A month later, we used our latest internal model that fully autonomously found this more complete version and very independently found the full version of what was released. The thing to realize about the AI models is that fundamentally, they're empowering tools. and that they're able to work across a lot more context than people are. You know, I myself actually have found a security vulnerability way back in 2010 in an open source piece of software. And I remember finding a small function that looked like it wasn't quite implemented correctly, but all of the Usages of it had been protected, but then actually someone added some code later that didn't protect in quite the right way. And so it's this kind of sleuthing that you do as a person. But it's very hard to keep all that context in your head, to have expertise across many different disciplines, and to think about how the software might be used in practice. And so that capability, which is what makes AI so useful, that it means that it can help us solve problems, it's also something that helps us find flaws in the software that people have already written. 00:02:24 Speaker 2: And Greg, as a follow up, a specific question I got for you is how much of the work in this case study was literally done by AI versus a researcher? 00:02:34 Speaker 3: So for the Calif side, I'd say, I don't know how to put an exact percentage on it, but the researcher and their deep expertise, let's call it 75%, 80%. Whereas for the OpenAI internal version, that was really, we set up an environment and we took our most capable model with a lot of compute behind it. And that autonomously was able to find the full exploit. 00:02:56 Speaker 2: Matthew, I want to go to you on probably the scarier bit, which was reading the blogs and the reporting. It seems like hackers tried to exploit these flaws before they were made public, right? Which raises the question whether AI was what found them first. Do you see what I mean? 00:03:13 Speaker 3: Absolutely. 00:03:13 Speaker 1: And I think what we have to be clear about here is that AI is the real hero of this story. These flaws have existed for quite some time. They're extremely serious. And PHP, the underlying software that contained the flaw, is present and running on something like 70% of the world's websites. 00:03:32 Speaker 3: So this was very significant data. And it was a very significant flaw. 00:03:35 Speaker 1: It's hard to know whether hackers, nation states otherwise, had access to this flaw that had been present for years. And I think that what OpenAI did with their tools and what the researchers did was extremely responsible. They found a latent bug that was there. They found something that may have been exploited for many years. They responsibly disclosed it to the industry and companies like Cloudflare. And we worked tirelessly behind the scenes to make sure that our customers were protected and that the Internet was protected before it got out. But it's clear it wasn't that AI created the bug. AI found the bug that was there. And it's incredibly important that we've now found it, fixed it, and the software is significantly more secure because of the work that OpenAI did. 00:04:23 Speaker 2: Greg, to the best of your knowledge, are there any examples of hackers using OpenAI's models to try and exploit this specific PHP flaw? 00:04:34 Speaker 3: So we believe that attackers are going to start using models broadly to try to exploit these flaws. We have monitoring on our side watching for it. And I think we're starting to see attempts or we have seen attempts in the past, but not at a massive scale yet. And I believe that we're going to see just a ramp up. I think that Cloudflare has reported seeing this kind of very steep ramp of attempted exploitation. So I think that we should expect attackers to try. But as Matthew said, that we have put in a lot of effort here to do responsible disclosure, to repair the internet and to make this software more secure. 00:05:13 Speaker 2: Matthew, same question to you, sort of knowledge or awareness of any specific incidents where open AI models have been used or an attempt to be made by hackers to use them. But also kind of give me the backstory of when Cloudflare came into this, when you got involved. 00:05:28 Speaker 1: So first of all, once the exploit is known, AI doesn't have to be used at all. You can actually, in this case, with a couple of requests to a web server, take over that web server. So once we know what the problem is, then that can be exploited without using. 00:05:45 Speaker 3: AI at all. 00:05:45 Speaker 1: And hackers right now are exploiting this flaw at just incredible rates. Literally, we're seeing millions of attempted attacks using this flaw. Now, across all of our network, every single minute of every single day. So the flaw is now known. And if you are running PHP on your servers, you need to update PHP to the latest version. 00:06:11 Speaker 3: It's free. 00:06:11 Speaker 1: You can get it updated immediately. And that's the only thing that's going to keep you safe from this. What I think, again, was incredibly responsible of the OpenAI team, and I think most of the frontier labs are working with companies like Cloudflare in this way, is that when they found this flaw a couple months ago, they reached out to us and they said, we found this. They disclosed it to us in a confidential way. We were able to work with our team to make sure that we had put protections in place so that when this was actually released, as much of the internet was protected as possible. And they didn't just work with Cloudflare, they worked with WordPress, with the PHP community, anyone that could actually make as much effort to protect this. And again, I think it's really important to say here, AI is the hero here. This is a latent flaw that has been in the code for a long time. It may have been being exposed by and used by nation states or other hackers to do all kinds of bad things. 00:07:10 Speaker 3: But now that it's been exposed, we've actually fixed it. 00:07:13 Speaker 1: And what I'm optimistic about is, yeah, there may be times when bad guys use AI to find things. But right now, we've got the good guys who have more resources, have access to great tools like the ones provided by OpenAI and the ones provided by companies like Anthropic. And we're going through code and we're going to find a bunch of flaws. 00:07:31 Speaker 3: We're going to find a bunch of bugs. 00:07:32 Speaker 1: But I actually think the Internet is getting more secure because of the work that companies like OpenAI are doing. 00:07:39 Speaker 2: Greg, you wrote quite an offensive post on X, sort of recalling your own history with PHP as a programming language. One of the ideas that you talked about was suggesting that the governments or I think you call them authorized defenders, but they scan the internet for vulnerable systems and in some cases just remotely patch them. And I think that raises a lot of questions about permission, like who decides that that should be the case and where the line gets drawn. 00:08:06 Speaker 3: So I think that we are... You should look at PHP as one example of the kinds of flaws that AI is now able to find. And we're in this world where there's what we call a defender's window, where frontier models are ahead of what's broadly available. And so that means that the labs, those who have access to these cutting edge capabilities can find vulnerabilities, can peer into the future before anyone else does. And so that means if you are a company and you secure yourself using the frontier models, you will be more secure and able to stay ahead of what is broadly available and what attackers can do. And so I think that defender's window is a very important concept for every company, for every defender to really think through. But it's also very clear we're going to be finding lots and lots of flaws that are there in the software, that have been there for many years, that are tech debt, that have piled up as people have built software on top of just many, many years' worth of accumulated tech debt. And so to me, the question is, how does the internet stay secure? And there is precedent. With, for example, as botnets became prevalent, having the FBI get court orders to clear out malware from botnets because that's something that's important. It's not just the site owner who is affected if they get attacked. It's really the whole internet becomes less robust and secure if there isn't someone who's trying to make sure that patches are applied or that if there is sort of malicious activity, that there is some sort of way of notifying site owners and making sure that improvements are made. And so I think that the core question we have, I don't think that we have the answers yet, but I think that we're really asking the question is, as the scale of vulnerabilities found increases, and it's harder and harder for anyone to make sure that they're taking all of the patches, how does this whole system need to change? And so if you look at PHP, we had two months worth of very manual process. We were on the phone with many of our partners, the government to really try to land this well. We see other vulnerabilities coming. We've been working on several of them as well. And I think that the system of making sure that we stay ahead, that defenders are advantaged, and that there is some sort of process, some sort of societal process for keeping the internet secure. That's the question that we think that the people should be asking. 00:10:28 Speaker 2: Greg, before I let you go, I've been using DOTS a lot alongside Muse, Grokbot. I kind of put this question to Sam the other day, trying to work out how to make an agent useful to me. Is there any updated data you can give me on how DOTS has gone from an adoption perspective and a user perspective? 00:10:48 Speaker 3: It's really hockey sticking. I mean, it's really exciting to see all the usages. And actually, I think that there's so many surprising things across work and also your personal life. And even, for example, for protecting people, one of my friends told me that Her dot noticed a phishing email in her inbox, said, hey, do not open this email. It's a phishing attempt. And her admin also has access to her email, and it also pinged her admin saying, do not open this email. And so I think that we're finding this proactive form factor of an AI that's looking out for you in your personal life, in your work life, to be something that just increases people's helps people save money, is able to increase people's productivity. And it's something you could not get a person to do, right? To be looking out for you constantly. So I think that there's just so much potential that's being unlocked. And I think that we can make people more productive, make work more joyful, and also help people just live more fulfilling lives. 00:11:51 Speaker 2: Greg Brockman of OpenAI, Matthew Prince, CEO of Cloudflare, Cloudflare key to protecting a lot of what's happening in internet infrastructure. Thank you both very much.