SGAIL Labs.

It started as security infrastructure. The first work was the plumbing that autonomous systems need and rarely have: a firewall for AI-to-AI traffic, a tamper-evident witness log, and a family of detectors for prompt injection, obfuscated input and escalation. That work is still here, on the Infrastructure page.

Building detectors meant collecting failures. And the failures that mattered kept turning out to be the ones nobody had written down. They weren't exotic attacks. They were the situations a working electrician, plumber or landlord recognises on sight — the confident wrong diagnosis, the reading from the wrong sensor, the tenant story that fits every fact but one — and that no benchmark asks about, because a benchmark asks questions and these are situations.

So we started writing them down with the people who were there. That became the Rubric Catalog. A situation needs somewhere to happen, with a business around it and consequences after it, so the catalog got a town: Raccoon Ridge. Watching AI agents talk to each other surfaced failures nobody had a record for yet, which became the Hive Mind. And a finding that can't change anything is only half useful, which is what the Training Matrix is for.

The result is a lab with a firewall, a contributor wall of tradespeople, a simulated town and a machine-mined corpus under one roof. It looks unusual because the problem is: testing AI where it has to actually do something draws on all of it.

Each page on this site marks what is built, what is in development and what is experimental. We are pursuing DHS SBIR funding to accelerate the work, while continuing to take on evaluation, training and research engagements. If any of that overlaps with what you're building, we'd like to hear from you.