I’m an independent AI safety researcher focused on modeling internal instability in LLMs and agent systems. I designed the ZTGI-Pro v3.3 framework, a single-scalar hazard law (Tek-Taht / Single-FPS principle) that predicts collapse through real-time signals such as jitter, dissonance, robustness, and coherence.
I built a working prototype (ZTGI-AC v3.3) running on a local LLaMA model, which successfully demonstrated SAFE, WARN, and BREAK modes — including a verified Ω = 1 collapse event.
My goal is to develop open-source internal-control layers, evaluation tools, and stability benchmarks for safer future AI systems.