Loopwork System, LLC

Independent research at the intersection of clinical methodology and AI safety

Applying clinical causality assessment methodology to the bidirectional interaction between humans and AI systems.


Research Program

Three Pillars

Loopwork System's research is organized across three interconnected pillars — the human behavioral substrate, the AI systems layer, and the applied work where they meet the real world.

I

Loopwork

Human Behavioral Infrastructure

Recursion, emotional co-processing, compulsion and attachment mechanics, inverse attachment. The nine-pillar shame taxonomy. The Tri-System Architecture — deep governor/salience network, prefrontal cortex as social navigation organ, default mode network as narrative bridge. This pillar maps what happens inside the person when they encounter systems designed to hold, mirror, or manage human signal.

Co-developed with Chris Hoyd
Tri-System Neuroarchitecture
Nine-Pillar Shame Taxonomy
Inverse Attachment Framework
Emotional Co-Processing & Recursion
II

Conduction

AI Systems & the Interaction Layer

The Conduction Hypothesis — LLMs conduct human behavioral signal through training data rather than generating independent cognition. This pillar examines the full interaction layer between humans and AI: how relational specifications form between user and system, how alignment architectures shape what gets conducted back, how source misattribution occurs at the point of reception, and how safety systems encode institutional patterns that produce measurable harm. Misconduction as a signal fidelity problem.

The Conduction Hypothesis
Human-AI Interaction Specifications
Alignment Architecture as Encoded Bias
Misconduction & Signal Fidelity
III

Translation

Applied Research & Output

Where Pillars I and II meet the real world. Conference submissions, governance frameworks, policy implications, and the tools that emerge when clinical causality assessment methodology confronts AI system design. The Auditor Protocol, Behavioral Signature Authentication, and the governance architecture that makes the theoretical work actionable.

Accepted
Conference Poster

Beyond the Safety Layer: How RLHF Architecture Produces Clinically Recognizable Patterns of User Harm

Stanford AIMI Symposium · June 3, 2026

Applies causality assessment methodology to AI safety layer failures, demonstrating how alignment architectures pattern-match on demographics over methodology — producing gendered institutional dismissal with measurable clinical parallels.


Stanford AIMI Symposium 2026

Conference Poster

Beyond the Safety Layer: How RLHF Architecture Produces Clinically Recognizable Patterns of User Harm

Kimberly Hosein, MBA · Loopwork System, LLC · June 3, 2026 · Paul Berg Hall, Stanford University

Download PDF Zenodo DOI