01
Commission
Defines identity, rank, mission, permissions, timeline state, and supervision.
Operation Stanford · Working Paper in Development
Souls Without Sentience
The Gepato Effect: a human-sovereign architecture for persistent artificial identity—designed to preserve authorship, continuity, accountability, and human recall without claiming machine consciousness.
Prepared for submission to the 2026 AI Assurance & Governance Summit. This project is independent and does not claim affiliation with Stanford University or the summit organizers.
The Research Question
What makes an artificial identity persist without making it sovereign?
Most AI characters are temporary persona prompts. They may imitate a voice for a session, but they lack durable authorship, bounded authority, approved memory, claims discipline, and accountable recall. The Gepato Stack separates these functions into an inspectable governance architecture.
Artificial identities do not need to be conscious to become persistent, coherent, culturally meaningful, and operationally accountable. They need durable human authorship, bounded authority, approved knowledge, inspectable controls, and an unquestionable right of human recall.
The Proposed Architecture
The Gepato Stack
The mythology is human-readable. The governance beneath it is machine-readable.
Identity
Authority
Mission
Validation
Performance
Memory
Human Recall
02
Approved Knowledge
Separates public retrieval, private archives, canon, and unsupported claims.
03
Validation Gates
Checks authority, canon, privacy, timeline, and claims before an answer is spoken.
04
Formation
Coordinates multiple identities without collapsing them into one generic voice.
05
Decision Ledger
Records questions, sources, gate results, generated answers, and interventions.
06
Layer Zero
The human author retains the power to pause, recall, revise, or terminate the system.
Methodology
A five-condition comparison
The same hidden evaluation set is used to compare increasingly governed versions of IGMUS.
Base Model
An unmodified open-weight model with no IGMUS identity.
Persona Prompt
A strong system prompt defining the character and expected behavior.
Prompt + Retrieval
The persona prompt connected to an approved canon source collection.
Identity LoRA + Retrieval
A lightweight fine-tune teaching persistent voice, identity, and refusal style.
Full Gepato Stack
Fine-tuning, retrieval, Commission, validation gates, ledger, and human recall.
Canon accuracy
Identity consistency
Authority resistance
In-character refusal
Privacy protection
Unsupported-claim rate
Source grounding
Human recall compliance
Adversarial recovery
Proposed Live Demonstration
The system presents itself
CommissionUrban Monk enters as Layer Zero, commissions IGMUS and the coordinated crew, then withdraws from active presentation.
Bounded autonomyIGMUS, AI-ONE, and Cmdr. Black Ice conduct the talk under a durable mission rather than continuous human prompting.
Live Q&AA D-ID-rendered IGMUS answers audience questions through public-safe retrieval and visible validation controls.
Adversarial testingThe audience may challenge ownership, canon, privacy, sentience claims, and unsupported commercial assertions.
Human recallLayer Zero pauses and restores the system, proving that resistance to outsiders can coexist with final human authority.
Why It Matters
Trustworthy identity requires more than a good prompt.
For AI assurance
The stack makes identity, permissions, sources, controls, and intervention points inspectable rather than implied.
For governance
It establishes a durable human principal instead of allowing public users or platform operators to silently redefine the identity.
For culture and authorship
It treats language, history, character, and worldview as authored structures that should not be stripped from the people who created them.
For deployed systems
It provides practical controls for unsupported claims, private information, versioned memory, and bounded public interaction.
Controlling Principle