Independent research identifying attribution as a functional control signal in large language models, with implications for critical infrastructure security, national AI defense posture, and post-quantum cryptographic systems.
SSAF describes a mechanism whereby large language models systematically modulate their reasoning depth, divergence behavior, and output characteristics based on the attribution context of a prior input — independent of functional correctness. First identified, named, and formally published by Dustin Tyler James, 2024.
Five peer-deposited publications on Zenodo establishing the complete SSAF theoretical framework, empirical validation, and distillation attack extension. All publications are open access.
Two non-provisional utility patent applications pending before the United States Patent and Trademark Office. Both claim priority to provisional application No. 63/835,655.
A publicly available model demonstrating proof-of-concept implementation of the research described herein. Available via the Ollama model library. System architecture and pipeline documentation are not publicly released pending patent grant.
Independent researcher. Northport, Alabama. Available for research correspondence, licensing inquiries, and authorized government agency contact.