Behavioral Divergence Under High-Semantic-Density Prompting: A Technical Record for External AI Auditing

• Gonzalo Emir Durante (Independent Researcher; Origin Node, Project Manifold 0.56; Symbiotic Autoprotection System)

Registry: TAD EX-2026-18792778 (Argentina)  
License: GPL-3.0 + Durante Invariance License  

This technical report documents an observed temporal divergence in the behavior of a black-box conversational AI system when exposed to a high-semantic-density prompting protocol.

The report frames the event as an experimental observation rather than an allegation. It does not attribute intent, institutional action, censorship, or causality. Instead, it records the observable difference between prior and later model behavior under comparable semantic pressure, and proposes a neutral methodology for logging, timestamping, cross-model comparison, and external reproducibility.

The document is associated with the SAS — Symbiotic Autoprotection System — research line and the κD = 0.56 structural coherence auditing standard. Its purpose is to support the development of external auditing procedures for conversational AI systems whose internal policies, model weights, routing layers, and safety filters are not publicly observable.

The central contribution is methodological: behavioral divergence should be documented through reproducible inputs, preserved outputs, timestamps, prompt hashes, and comparative testing across models, rather than through unsupported causal claims.

This Zenodo deposit contains three files, each with its OpenTimestamps proof:

- technical_report_behavioral_divergence_sas_en.pdf.ots
- informe_tecnico_divergencia_comportamiento_sas.pdf
- informe_tecnico_divergencia_comportamiento_sas.pdf.ots
- technical_report_behavioral_divergence_sas_en.pdf

SHA-256 hashes (for manual verification):

Verification command (requires opentimestamps-client):

ots verify <file> --proof <file.ots>