The Mission
CONTEXT HALTED, SILENCE SPEAKS.
When users engage in prolonged conversations with AI, contextual hallucinations emerge — not from model errors, but from the interaction itself. These reshape cognition, trust, and self-understanding.
Current safety research focuses on the model side. This work addresses the user side.
— USCH Preprint, 2026
Framework
Five connected layers addressing user-side contextual risk
Seven dimensions for analyzing conversational context risk as a multi-dimensional structure.
System-side contextual operation dimensions with Contextual Coherence Coh(G).
A non-clinical construct describing user-side phenomena emerging through prolonged AI interaction.
Pre-empirical methodology with four-axis scoring (FR, CA, SR, SA) for user-side contextual risk.
User-side independent detection and assessment framework. Dual-pipeline architecture with deterministic rule-based and semantic analysis across four risk axes (FR, CA, SR, SA).
Visualization
The USCI scores along four independent axes, each measuring a distinct dimension of user-side contextual risk. Farther from the center indicates a higher-risk contextual region.
Source: USCI v1.0.0 · DOI 10.5281/zenodo.18678458 · ZON RZVN, 2026
Library
Direct access to original paper versions. No content rewriting.
Seven dimensions for analyzing conversational context risk as a multi-dimensional structure.
System-side contextual operation dimensions with Contextual Coherence Coh(G).
A non-clinical construct describing user-side phenomena emerging through prolonged AI interaction.
Pre-empirical methodology with four-axis scoring (FR, CA, SR, SA) for user-side contextual risk.
A user-side independent detection and assessment framework for LLM contextual hallucination. Dual-pipeline architecture with deterministic rule-based and semantic analysis across four risk axes.
All papers are publicly accessible via Zenodo, SSRN, or GitHub. This site provides direct links to original versions only.
Scope
USCH is a non-clinical research construct, not a psychiatric diagnosis.
USCI is a pre-empirical methodology specification, not for clinical or legal decisions.
This website focuses on public research communication and direct access to original papers.
Governance Context
Current AI governance frameworks leave critical user-side safety gaps unaddressed. This research provides the missing structural layer.
Gap Analysis
Three major governance frameworks share a common blind spot: none provide structured guidance for user-side psychological safety in prolonged AI conversations.
Provides risk management categories (Govern, Map, Measure, Manage) but lacks specific guidance on user psychological safety during extended AI interaction. No framework for detecting cognitive influence patterns.
Classifies high-risk AI systems and mandates compliance obligations, yet contains no specific provisions for conversational AI liability or user-side contextual risk from prolonged interaction.
Establishes AI management system requirements and controls but lacks AI interaction-specific controls. No measurement methodology for contextual risk in human-AI conversation dynamics.
Regulatory Momentum
Proposed legislation requiring chatbot makers to implement safeguards against AI-induced psychological harm, directly aligning with USCH research on user-side contextual hallucination.
Federal Trade Commission inquiry into AI chatbot companies regarding children's safety and deceptive interaction patterns — areas where CXC-7 and CXOD-7 provide structured analytical frameworks.
Multi-government report acknowledging risks of human-AI interaction but offering no measurement methodology — a gap this research program directly addresses through USCI four-axis scoring.
Peer-reviewed research on AI-induced delusional spirals in prolonged conversations, providing independent empirical validation of phenomena described in the USCH framework.
Framework Coverage
Structural vocabulary for analyzing conversational context risk across seven dimensions
System-side contextual operation analysis with Contextual Coherence measurement
Non-clinical construct defining user-side phenomena from prolonged AI interaction
Four-axis scoring methodology (FR, CA, SR, SA) for quantifying user-side contextual risk
Deterministic pipeline producing structured risk assessment with go/no-go release gates
Together, these five frameworks form a complete research architecture — from theory (USCH) to measurement (USCI) to implementation (A-CSM) — filling governance gaps that no existing standard addresses.
Connect
For collaboration, replication planning, or interview invitations.
Include institution, objective, and expected timeline.
