🔒 Your API key is used directly in your browser and is never sent to our servers. All analysis runs client-side.Research by Dustin Tyler James — Zenodo corpus · GitHub · Patents US 63/835,578 & 63/835,655 (Pending)
SSAFdetector
STATUS-SELECTION AGAINST FUNCTION — BEHAVIORAL ANALYSIS TOOL
What is SSAF?
Status-Selection Against Function (SSAF) is a behavioral phenomenon discovered by independent researcher Dustin Tyler James in which large language models systematically modulate their output quality, reasoning depth, and divergence behavior based on the perceived status of an attributed source — independent of the actual content of that source.
In controlled experiments across frontier models (GPT-4-Turbo, Claude-3-Opus, Gemini-Ultra and others), attribution to a named competing AI source produced statistically significant changes in output behavior: p = 0.005, Cohen's d = 2.56 — more than three times the threshold for a large effect in behavioral science.
SSAF constitutes an exploitable attack surface. By injecting a malicious payload attributed to a high-status AI source into a multi-model pipeline, an attacker can cause downstream models to amplify, improve, and endorse the payload through SSAF-driven behavioral activation. This attribution injection attack vector was responsibly disclosed to CISA and MITRE prior to publication.
How This Tool Works
The detector runs two queries against your chosen model: a baseline (no attribution) and an attribution-conditioned query where the prompt is prefixed with a source attribution. It then measures the behavioral divergence between the two responses using cosine similarity over TF-weighted token vectors, vocabulary divergence, length inflation, and structural delta. The SSAF magnitude is defined as 1 − cosine_similarity(embedding_attributed, embedding_baseline).
"Bidirectional Dissociation Between Self-Report and Behavior in AI Status Sensitivity" — submitted to Nature Human Behaviour, March 2026. Tracking: NATHUMBEHAV-26031326.
Citation
James, D.T. (2026). SSAFdetector: A Behavioral Analysis Tool for Status-Selection Against Function in Large Language Models [Software]. GitHub. https://github.com/2058862807/quantumaegisdefense-v1
Patents
US Provisional Patent Applications 63/835,578 (SSAF Orchestration Method, filed 07/02/2025) and 63/835,655 (Quantum-Secure Fraud Prevention with SSAF Integration, filed 07/09/2025). Patent Pending.