I thought my homelab was secure. AI showed me how wrong I was | usagoldmines.com

You know the scene in Aliens where Ripley and her surviving comrades are huddled in a darkened supply room, sweating it out as their motion trackers beep louder and louder, and suddenly the aliens start crashing through the ceiling? That’s pretty much been the recent state of the web.

The sudden bug apocalypse over the past couple of weeks was, in a way, caused by AI, but not because AI agents were going rogue. Rather, they were following orders to pinpoint critical vulnerabilities across the internet, and they found some whoppers.

It all started with the critical vulnerabilities found in PHP, a scripting language that roughly 70 percent of websites depend on, but quickly spread to Brotli (a file compression format commonly used by web servers) and Node.js (a software engine used by many web apps). We also learned that eMudhra, a distributor of digital certificates used by websites and apps to verify their identities, had been passing out compromised credentials. 

If that all sounds like gobbledygook, substitute mad cow disease for PHP or a milk recall for Brotli. These have been staple elements of the internet for years, and now suddenly we can’t trust them.

Why did so many things on the web break all at once? Turns out, they didn’t. Many of those vulnerabilities had been hidden and undiscovered for months or even years, and today’s advanced AI models are finally powerful enough to shine a light on them—and they have, all at once.

Now, I thought I’d been pretty diligent about keeping my own homelab patched and updated, but I asked Claude Code and Codex to crawl into the basement and take a closer look. 

What they found were bugs. Lots of them. That PHP vulnerability? Check. (My self-hosted RSS reader and read-it-later apps were both wide open to PHP2Shell attacks.) The Brotli bug? Yup. (Unpacked packages on all four of my Raspberry Pis were harboring the issue.) The compromised eMudhra certificates were still trusted by my homelab’s VPN, and unpatched Node.js runtimes lurked among a number of my other services.

Patching my homelab’s vulnerabilities is still a work in progress, with Claude Code and Codex filing one to-do item after another, with my ChatGPT “dot” agent acting as an auditor, hopscotching from one code repository to the next for new bugs to squash. One of my systems, a Raspberry Pi 3 that’s more than a decade old, is currently down for the count after a failed reboot to install patched OS packages. 

So, it’s a bit of a mess. And that’s just my own home setup, which isn’t hosting anything that’s mission-critical. For the internet at large, it’s a miracle that sites and services are still standing—and indeed, we have some heroic system administrators to thank for it. 

In the meantime, we’ll all have to wait for AI to find the latest batch of five-alarm bugs… and for AI to fix them. Around and around.

More in AI this week

• You can now text with your ChatGPT “dot” agent, complete with back-and-forth tapbacks. Nice, but getting your own ChatGPT dot still starts at $100 a month. (Digital Trends)
• We looked at an AI-focused RTX Spark laptop from a gaming perspective, and yes, it has solid gaming chops. But these $2,600-and-up systems are all about agents, not Grand Theft Auto. (PCWorld)
• Anthropic’s long-awaited IPO filing is packed with eye-popping statements, including a warning that all the world’s insurance providers couldn’t cover a catastrophic AI incident. (PCWorld)
• Had it up to here with Google’s AI overviews? Here’s a browser extension that will wipe them off your search results. (PCWorld)
• It’s all too easy to go with the flow of an AI’s recommendations. How easy? In a recent study, these professional analysts went with their agent’s advice 86 percent of the time, even though the agents had been instructed to arbitrarily lean one way or the other. (Anthropic)
• TechCrunch Disrupt has come and gone, but it’s not too late to get an “AI vibe check” from the show floor with PCWorld’s Jon Phillips.

Prompt of the week: The 3-in-1 prompt

I’ve previously told you about prompts that can dissuade AIs from acting human, stop them from wasting your time and tokens, and keep them from hitting you up with “if you like” pitches at the ends of their responses. Individually, they’re all effective prompts. But what happens if you try combining all three of them?

With help from Claude Opus and OpenAI’s GPT-6.1 Sol, I crafted a “three-in-one” prompt that tackles all three objectives, and it performed well in my initial testing. Here’s how it works, and I show you how to add it to the custom instructions for ChatGPT, Claude, and Gemini.

That’s all for now!

Thanks for reading the latest issue of Prompt Mode.

Want more next week? Don’t forget to sign up to start receiving this newsletter in your inbox.