Will AI expose the unstable infrastructure of the internet?

A surprising announcement was made earlier this week. A recently discovered vulnerability in the open source programming language PHP required website owners to patch their sites immediately. If we couldn’t update right away, we had to take down the site to be safe.

On this week’s episode of TechCrunch’s Equity podcast, Kirsten Kolosek, Sean O’Kane, and I discuss the significance of the PHP disaster and whether we’ll see more like it in the future.

Sean called this “a real moment of reckoning,” highlighting how “when you get down to the bones of the internet,” everything seems to be “held together by glue and dreams.” And with widespread access to large-scale language models and AI agents, discovering these vulnerabilities will become even easier. (That said, these particular vulnerabilities were discovered by human researchers, not AI models.)

Kirsten, on the other hand, wondered: “Are we about to see the next level of prepper doom?” “Because we see how banks can be exploited. And (there are) data and privacy concerns that I feel are much bigger than we can solve right now.”

Continue reading for a preview of the conversation, edited for length and clarity.

Sean: The internet seems to be broken in more ways than one this week.

I think this is the best thing to think about when people are facing this kind of cold. How many times over the past decade have we encountered a moment where something broke, something was about to break, and one person maintaining one small repro repository as a hobby realized something nefarious was going on and saved us from a complete collapse of the Internet’s infrastructure?

Multiply this by a billion and you not only have access to very advanced software in the form of LLM that everyone uses, but you also have the ability to create AI agents dedicated to accomplishing those things and tell them to do it themselves.

There are two balancing events happening this week. One is that a lot of labs and companies like Google and OpenAI have actually sent people out to discover a lot of vulnerabilities, including one in PHP. And many people seem to be rushing to exploit those vulnerabilities to hack into websites. As a result, governments are shutting down websites, and some very popular websites are temporarily down to fix this vulnerability.

This is a moment of truth, and the downside risks of building much of the Internet with open source technologies like PHP will become clear. When you get down to the bones of the internet, it is held together by glue and dreams.

Anthony: There were a lot of things that stood out to me about this. One is the fact that this is not a new vulnerability, at least as I understand it. This is a vulnerability that has existed in PHP for decades.

And since PHP has been used as the programming language for building much of the web, this vulnerability has been around for decades, but no one knew about it. Essentially what happened is that these big AI companies (including Wiz, which is now part of Google) were able to use LLM to discover these vulnerabilities. (As mentioned above, I misunderstood this. Human researchers participating in OpenAI and Wiz’s Hack the Planet initiative discovered the vulnerability, but they were using AI to exploit the bug.)

I think part of the calculation is that if we can find this vulnerability using LLM, others can find it too, so we should patch this. However, the only way to release a patch is to publicize the vulnerability. This creates a very panicked situation. “This vulnerability has been around for decades. It needs to be patched now. If you can’t patch it now, shut down your website. That’s how serious the vulnerability is.”

Kirsten: I think the important thing is that a lot of people don’t even realize that their website has PHP in it. PHP, as you said, is an open source programming language that’s been around for decades.

For those who don’t know, hypertext preprocessor is the official term and the core element behind content management systems used by things like WordPress, Drupal, and 70% of websites. We’re talking about tens of millions of websites. For me, it’s the scale. It also immediately made me think of two things. One, this is just the beginning. What other decades-old malicious software could be exploited?

And the other thing that 404 Media mentioned, which had a really big breakdown, is link rot. Because a lot of people, businesses, governments, maybe small websites in your neighborhood are shutting down these (websites). This can lead to a lot of link rot and websites with perhaps small bits of history remaining inaccessible rather than being patched.

Sean: Anthony, I think you made a really good point. I’ve seen a lot of people complaining about how the disclosures have unfolded here, but something like OpenAI did with a post about it made me a little excited.

It also appears that several people got ahead of the embargo on this release, sparking competition among hacker groups. I think it was Matthew Prince, CEO of Cloudflare. I’ve already posted about how the PHP2Shell hack escalated into millions of attempts.

The problem with these vulnerabilities is how this typically works in cybersecurity. If you’re stress testing a company’s software, even if it’s basic software that helps run different computers or parts of the Internet or whatever, you can go to that company and publish it and stay in your bubble. But when you’re talking about something this deep and fundamental, you can’t go around and tell everyone carefully.

So if AI is advanced enough that you can command a billion agents to send an email to every individual who has control over some part of your website, you might be able to do it. We now regret the days when there were easily identifiable webmasters, but that might have solved the problem to some extent. But there’s no easy way to do this. I think this is a lot of back and forth and clearly the beginning of a new frontier in cybersecurity.

Kirsten: Matthew Prince’s social media post meant so much to me and was so scary. Also, I don’t know if you two saw this, but Amazon also published a blog post. This company is very well capitalized, so they have the tools and the team to figure out what’s going on. They essentially operated a honeypot network, or decoy, and saw attempts to exploit the vulnerability before it was actually made public.

For me, it comes back to this again. Amazon has the tools, capital, and know-how to combat these issues. Too many websites, even government websites, don’t get that much attention. And for me, that’s the really scary part.

And there are (there are) AI employees who are probably trying to take this to the next level, like Prepper Doom. Because you can see how banks can be abused. And there are data and privacy concerns that I feel are much bigger than anything we can fix right now. So I’m interested in two things. One is how extreme people’s behavior will be, but I also wonder if this will result in some new startups or open source tools (from companies like Nvidia) to create these layers of security.