A bit about myself, and my mission.

I spent eleven years on a tree crew and stayed on the ground for all of it. On the ground I was the rigging specialist. I kept the climber safe if something let go above me, and I ran the timing, getting branches and logs of every size and shape to come down safely and fit into places they ordinarily would not, which took finesse and planning. Most of it ran on silent visual contact with the climber up in the tree. A lot of what we did was like a synchronized dance, a ballet of physics where a mistake has real consequences. We made very few mistakes over eleven years, and my father did the same work for twenty years before me. My family runs back through generations of broken people, and in the house I grew up in the damage never got a name. I picked it up young from the adults around me. The tree work was my day job. At night I built tools in my free time. I cheated at games and reversed them, working through their memory hooks and offsets to learn how the systems ran from outside. I still work on AI and its governance from outside the usual circles. There is little traction, and few people outside it know the work is happening. It is the same method I brought to the games, coming at a system from the outside and pushing it to see how it behaves. I build tools that attack AI systems, and I use AI to build them. That is the purple team method. I aim it at AI safety and alignment. I build the adversarial side so the defensive side has something real to test against, then I feed what the attacks find back into the defenses that are supposed to hold. That is threat-informed defense. The method sits close to me because it runs along my own impulsiveness and the line where accountability starts, and because I still falter. The companies building systems at scale will not hold themselves to account. I mean the class running the large platforms and the feeds that decide what a lot of people see. I watched what the platforms did to people's attention and to what they take as real. Agentic AI is the next system like that, and the check on it has to come from outside. One thing I finished on my own is a rendering preset called E L D E R . E N B, built for Skyrim SE. I made it because I wanted a specific quality of light in the game that I could not get any other way, and I built it with no money coming in for the work. It went up on NexusMods, where the download count is approaching one million, with well over a hundred thousand separate accounts running it inside their game. I work under the name Zentropy Labs. It is not filed as anything and there is no payroll, because I have nobody else to pay. The tools I build under that name are aimed at frontier AI labs, and no pilot is in place at any of them yet. My main tool is called Flywheel, and it is accountability infrastructure for agentic AI. Every tool call the agent makes gets a sealed receipt attached to it, signed with ed25519. There is a false-accept corpus I test the verifier against, so I know what it will wave through wrong. A learning loop on top of that lets an organization watch its own drift over months. Any run the tool produces has to be re-derivable by somebody outside my machine. If it is not, it does not ship. I have a young son. He will grow up with AI already part of the world. Safety researchers keep resigning from the largest labs. They do not think the models are being made safe fast enough. The people who run those labs are now asking the industry to pace the frontier and put more into alignment and outside evaluation. That is the work I already do, and I want to keep doing it while he grows up.