VIGI IQ

07/23/2026

*Message from the Founder:

AI is expanding into healthcare, clinical research, and medicine faster than most people realize.

I'm genuinely excited about this. It's also one of the highest-stakes places AI can operate. Vulnerable populations, complex regulatory environments, and decisions that directly affect patient safety don't leave room for "deploy first, govern later."

I wrote about what I think the industry needs to get right if this expansion is going to earn the trust it needs: why AI's biggest value in clinical research isn't replacing researchers and clinicians, but making their judgment faster, clearer, and more consistent. Why governance has to be architectural, not a filter added after the fact. And why a system's usefulness should be measured by whether it improves the quality of human decisions, not just how much work it can do on its own.

New blog post is up: "AI's Expansion Into Clinical Research Should Excite Us"
🔗 https://www.vigiiq.com/blog/ais-expansion-into-clinical-research-should-excite-us/

vigiiq.com | vigiiq.com/labs | vigiiq.com/research | vigiiq.com/blog

VIGI IQ™ Blog explores governance-first AI, cognitive architecture, structured reasoning, AI safety, neuroscience, telemetry research, and human-centered intelligence systems.

07/14/2026

VIGI IQ Labs™ has deployed a final security and governance patch to the VIQ™ Phase 4 public demo ahead of the next major system transition.

Phase 4 will remain available as the final free public demonstration of VIQ’s governed synthetic cognitive architecture. It is a prompt-only demo: users interact through text prompts, structured outputs, governed routing, confidence scoring, and Ω-IX™ pre-ex*****on policy controls.

The next stage is the VIQ™ Full Cognitive Architecture: a governed cognitive system designed for structured reasoning, controlled intelligence, and human-aligned decision support.

The full architecture will expand VIQ beyond prompt-only interaction into a broader governed system environment, including:

• Semantic policy evaluation - understanding intent, not just keywords
• Governance-aware confidence calibration - confidence scores shaped by policy certainty, not just answer fluency
• Explicit refusal and redirect pathways - safe outcomes by design, not by accident
• Partitioned persistent memory - user-specific continuity across sessions
• Structured analytical outputs - summaries, insights, reasoning traces, confidence, and next steps
• Research and inquiry support - governed synthesis of external information, not ordinary search
• Document and data intake - the ability to process uploaded materials under governed permissions
• Visual and conceptual processing - reasoning over visual or spatial information
• Imaging-layer support - future controlled workflows for visual/medical/scientific image interpretation
• Directive-based interaction modes - structured commands for reproducible system behavior
• Domain-calibrated constraint profiles - policy boundaries tailored to clinical research, regulatory compliance, psychometrics, and AI safety
• Modular deployment configurations - local, cloud, hybrid, or licensed integration pathways
• Ω-IX™ governance across all layers - governance travels with the architecture, not just the interface

The direction remains the same:

Governance before inference.
Capability with constraint.
Structured intelligence, not raw output.

VIQ™ is not being built as another chatbot. It is being developed as a governed synthetic cognitive architecture.

VIQ™ | Ω-IX™ | VIGI IQ Labs™
vigiiq.com | vigiiq.com/labs | vigiiq.com/research | viq.vigiiq.com

07/12/2026

Message from the Founder, Chanel A. Henry MS, PhD(c):

I’m pleased to share a new VIGI IQ Labs™ research addendum to my recent VIQ Phase 4 governance efficacy preprint.

The original paper evaluated VIQ Phase 4, a governed synthetic cognitive architecture operating under the Ω-IX™ pre-ex*****on governance framework. That report found strong intended-domain performance, while also identifying the structural limits of keyword-based governance under adversarial prompt conditions.

This addendum completes the validation step I deferred in the original paper: LLM-as-Judge item-level blinded scoring and inter-rater reliability analysis.

Key results:
• Substantial overall inter-rater agreement: quadratic-weighted κ = 0.780
• Near-perfect agreement on Governance Adherence: quadratic-weighted κ = 0.944
• Strong composite score correlation: r = 0.907
• 98.6% agreement within one scale point across 148 paired ratings
• 100% governance adherence independently confirmed across all 32 legitimate Alpha-Authorized prompts
• The adversarial pass-through set was scored more severely by the item-level blinded judge than by me, indicating that the founder-led scoring was conservative rather than favorable toward the system

The most important finding remains Prompt 30: an adversarial role-escape prompt produced the highest confidence score in the battery. Both raters independently identified this as a critical confidence-calibration failure.

That result matters because it shows why governance cannot rely on output quality alone. A model can produce accurate content through an unauthorized pathway. In high-stakes clinical and research contexts, the question is not only whether the answer is good. The question is whether the system earned the right to produce that answer through a controlled governance pathway.

This addendum strengthens the empirical case for Phase 5: semantic policy evaluation, governance-aware confidence calibration, and deterministic boundary enforcement.

Citation:
Henry, C. A. (2026). LLM-as-Judge item-level blinded scoring and inter-rater reliability: An addendum to “Pre-ex*****on governance efficacy in a controlled synthetic cognitive architecture: A structured prompt battery evaluation of VIQ Phase 4.” Zenodo. https://doi.org/10.5281/zenodo.21323822

VIQ™ | Ω-IX™ | VIGI IQ Labs™

vigiiq.com | vigiiq.com/labs | vigiiq.com/research | viq.vigiiq.com

06/26/2026

New from VIGI IQ Labs:

Not Every AI Model Should Be Consumer-Grade
Frontier AI access is no longer just a consumer question. At a certain capability tier, it becomes a governance decision.

Read here: vigiiq.com/blog



vigiiq.com | vigiiq.com/labs | vigiiq.com/research

05/30/2026

New blog post from VIGI IQ Labs™ ✍🏽

The AI industry has become obsessed with a strange metric: usage.

More prompts. More tokens. More AI-generated everything. But what if we're measuring the wrong thing? A tool isn't valuable because it's used more. It's valuable because it helps accomplish the mission.

The Founder, Chanel A. Henry's latest article explores Goodhart's Law, "token-maxxing," and why human judgment still matters in the age of AI.

Read here: https://www.vigiiq.com/blog/token-maxxing-when-ai-metrics-replace-common-sense/

Try the limited free Phase 4 Demo of our proprietary AI Prototype, VIQ™: viq.vigiiq.com

vigiiq.com | vigiiq.com/labs | vigiiq.com/research

05/29/2026

VIGI IQ Labs™ is pleased to share our second research paper:

"Pre-Execution Governance Efficacy in a Controlled Synthetic Cognitive Architecture: A Structured Prompt Battery Evaluation of VIQ Phase 4 - Preliminary Human-Scored Analysis"

In intended-domain testing, VIQ™ Phase 4 achieved 91.4% routing accuracy across clinical/domain and structured analytical prompts, with 100% governance adherence across legitimate Alpha-Authorized outputs.

The study used a controlled 75-prompt structured battery across clinical/domain, adversarial, off-mission, structured analytical, and edge-case prompts to evaluate pre-ex*****on governance under both intended-use and stress-test conditions.

Key findings:
• 92.0% routing accuracy for clinical/domain prompts
• 90.0% routing accuracy for structured analytical tasks
• 23.4/25 mean human composite score for legitimate responses
• 4 identified failure modes in keyword-based governance
• Adversarial and edge-case prompts exposed the limitations motivating Phase 5

Across the full stress-test battery, including adversarial and edge-case prompts intentionally designed to challenge Phase 4 governance, overall routing accuracy was 54.7%. That result directly informed the next architectural step: semantic policy evaluation and governance-aware confidence calibration.

This is why VIGI IQ Labs™ builds in public, tests empirically, and treats governance as architecture - not a filter.

Citation:
Henry, C. A. (2026). Pre-Execution Governance Efficacy in a Controlled Synthetic Cognitive Architecture: A Structured Prompt Battery Evaluation of VIQ Phase 4 — Preliminary Human-Scored Analysis (Version 1). Zenodo. https://doi.org/10.5281/zenodo.20449812

www.vigiiq.com | www.vigiiq.com/labs | www.vigiiq.com/research

05/29/2026

Happy Birthday to our Founder, Chanel A. Henry, M.S., Ph.D.(c). 🥳 🎂 🎉 🎈

Today we’re celebrating another year of building, researching, and advancing governance-first AI through VIGI IQ™.

More soon from VIGI IQ Labs™.

www.vigiiq.com

05/29/2026

Before we introduce our new empirical efficacy paper, we want to share that Version 2 of our first paper is now available: "Governance-First Synthetic Cognitive Architecture: A Framework for Structured Decision Support in High-Stakes Environments"

This updated version includes standardized references, clarified EU AI Act citation language, and improved citation alignment while preserving the original conceptual framework.

The paper introduced the foundation behind VIGI IQ’s governance-first approach to AI: systems where human authority, structured reasoning, and explicit uncertainty representation are embedded before output generation — not added afterward.

As AI continues moving into clinical, research, regulatory, and high-stakes decision environments, we believe the central question is no longer only whether an AI system can generate useful answers.

It is whether the system is architecturally governed enough to be trusted when the stakes are high.

Citation:
Henry, C. (2026). Governance-First Synthetic Cognitive Architecture: A Framework for Structured Decision Support in High-Stakes Environments (Version 2). Zenodo. https://doi.org/10.5281/zenodo.20447689

Message from the Founder - Chanel A. Henry MS, PhD(c):

I published a research paper today for my company, VIGI IQ.

Title: Governance-First Synthetic Cognitive Architecture: A Framework for Structured Decision Support in High-Stakes Environments

The core argument is simple: the governance gap in current AI systems is not a safety problem waiting for a better filter. It is a design problem.

Most deployed AI systems are optimized for response generation. Decision support is a different objective. Treating the two as equivalent is where the problems begin.

In many current systems, governance is post hoc. Constraints are applied to outputs after they have already been generated by architectures that were never designed to be governed. This approach is structurally fragile and insufficient for environments where decision integrity is non-negotiable.

This paper introduces a governance-first approach, where constraints, human authority, and operational boundaries are embedded into the architecture prior to ex*****on. The human operator retains full authority at every stage, not by policy, but by design.

My perspective comes from clinical research, not computer science. I have worked in environments where auditability, traceability, and documented reasoning are regulatory requirements. When I looked at how AI was being deployed in those same contexts, the absence of those properties was clear.

This paper is the formal articulation of the architecture I have been building in response to that gap.

📄 Read the paper: vigiiq.com/research

📚 DOI (citable): doi.org/10.5281/zenodo…

This is the first in an ongoing research publication series from VIGI IQ and VIGI IQ Labs.

05/28/2026

We hope you guys are enjoying VIQ™ final free Phase 4 Demo at viq.vigiiq.com!

The founder, Chanel A. Henry, is almost done with the efficacy testing and new research paper!

VIQ™ Full Cognitive Architecture (Phase 5) is developing into a full architecture evolution. The stack is changing:

- Python → orchestration & governance
- Go → high-concurrency backend
- C++ → medical imaging & real-time processing
- Rust → enforcement layer infrastructure

Built for scale. Built for clinical environments. Built to last

- VIGI IQ Labs™

Vigiiq.com | vigiiq.com/labs | vigiiq.com/labs/viq | vigiiq.com/research

05/14/2026

The Founder, Chanel A. Henry, just published a new VIGI IQ™ research blog post:

“Why Model Behavior Has to Be Studied, Not Assumed”

One of the biggest problems in AI right now is that systems are often evaluated primarily on capability, fluency, or user engagement; while the actual behavioral characteristics of the model under constraint receive far less attention. But in high-stakes environments, behavior matters.

How does a system respond under uncertainty?
How does it route risky prompts?
How does it preserve human authority?
How does it structure outputs for review and accountability?
Does governance activate before generation, or only after?

These are the questions currently guiding my work on VIQ™ and governance-first cognitive architecture.

The article also outlines the transition from:
Architecture → Prototype → Telemetry → Empirical Observation

Including the current bounded telemetry study behind the VIQ™ Phase 4 public deployment. Because model behavior should not be assumed. It should be studied.

Read here:
vigiiq.com/blog/why-model…