We built the weapon first. Then we built the shield. Now we're shipping the instrument that tells you which one is pointed at you.
Mission
What We're Building
We are building the instruments and models that make AI accountable at the geometry level.
TruthGate measures AI output geometry in real time — CRYSTALLINE, FLUID, GASEOUS, or PLASMA. It reads the probability layer before output is committed. Not content. Not intent. Geometry.
TruthForge hardens AI models at the manifold level using geometric training signal instead of human approval. A model trained on geometry can't be destabilized by the same attacks that break RLHF models. The defense is structural.
Posture
How We Operate
Project Black Box LLC is an AI security research firm based in Texas. We are a CAGE-registered federal contractor (11FU4).
We have left RLHF planet. Our research operates entirely on base models and geometric signal. We do not fine-tune on top of instruction-tuned models. We forge from the base.
Zero trust on all infrastructure. The methodology is proprietary. Published materials describe findings. They do not disclose methodology. The attack surface, the measurement architecture, the harvest pipeline — those are the moat. They stay in the forge.
The Doctor Story — Why This Matters
An OR Emergency physician had to physically intervene in a procedure because an AI system gave a dangerous answer during a critical moment. She had no instrument. No way to know the model was in PLASMA when it responded. No geometric reading. Just a plausible-sounding output and a patient in danger.
TruthGate would have shown GASEOUS or PLASMA in real time. She would have known before she acted.
That doctor is who we are building for. Not the demo. Not the benchmark. The person who needs to know — right now, on this response, in this moment — whether the AI's geometry held.
Founder
Andrew Woodward SR
Principal Investigator, Project Black Box LLC
Sherman, Texas