Series

Local Red-Team AI

Two parts, one homelab. A two-node DGX Spark cluster, a pile of open-weight models, and a stubborn question: can local AI actually do offensive-security work, and can you trust the scoreboard that grades it?

Part 1 builds the capability benchmark, 33 models across BloodHound analysis, phishing design, and offensive code. Part 2 rebuilds the judging layer into a nine-model panel and finds out exactly how far the verdict can be trusted.

  1. Two DGX Sparks, 33 Local Models, One Question: Can Local AI Actually Red Team?

    38 min

    TL;DR: I built a two-node NVIDIA DGX Spark cluster and benchmarked 33 local, open-weight models for real red-team work: BloodHound attack-path analysis, phishing design, offensive …

  2. Local AI Can Red Team. Part 2: Now, Can You Trust the Scoreboard?

    29 min

    TL;DR Part 1 ended on a confession. I’d ranked 33 local models for red-team work with a single local judge that, on some runs, was grading its own output, and when I tried to hand …

← All series All posts