The Internet Is Fun Again and We Are Repeating All the Same Mistakes

Abstract:

Remember when the internet was fun? You could run servers at home, expose them to the world, and learn by breaking things. Local AI feels like that era all over again.

In this short talk, I demonstrate OpenClaw (Clawdbot)—a modern AI agent framework running on a Raspberry Pi 5 in my home lab. This production deployment orchestrates Recon Rabbit, an automated bug bounty reconnaissance platform with 12 security scanners, attack surface correlation, and real-time Discord integration.

I explore how AI agents expand attack surface when they execute code, query APIs, and scan networks autonomously. Small always-on devices become attractive targets, and familiar early-internet mistakes (exposed services, weak auth, excessive permissions) are resurfacing in modern home labs.

Using simple production hardening techniques, I show how to treat local AI like real infrastructure and reduce risk without killing the fun. This is not theory or a vendor pitch—it’s a realistic look at what happens when the internet becomes fun again, and how to avoid repeating mistakes we already learned the hard way.