Building a Local-AI Bug Bounty Lab — and Where I Learned Not to Trust It
I connected an AI-driven recon tool to a self-hosted LLM running on my own GPU, drove it from a Kali attack VM, and pointed it at a ClassicPress target on a segment I control. Cross-host inference, a VRAM wall, a clean run against the wrong target, and the lesson I keep relearning: a local model is a lead generator, not an oracle.