An open architecture for deterministic AI governance that separates policy enforcement from model behavior, enabling trustworthy deployment across providers.
An open architecture for deterministic AI governance that separates policy enforcement from model behavior, enabling trustworthy deployment across providers.
Project Details
Updated 07/03/26 · Edited by orgAI is becoming increasingly capable, but the mechanisms for governing its behavior remain fragmented, model-specific and largly opaque. As organizations adopt AI across workflows, they need infrastructure that makes AI behavior predictable, auditable, and accountable - without depending upon the behavior of any single model.
This projects develops the Governed Intelligence Architecture Specification (GIAS), and open architectural framework for deterministic AI governance. Rather than attempting to change how foundational models reason internally, GIAS defines how identity, policy enforcement, execution, observation, persistence, and forensic replay should operate outside the model, allowing governance to remain consistent regardless of which AI provider is used.
Funding would support continued development of the reference architecture, implementation guidance, and the working reference implementations that demonstrate the spec in practice. This includes the continues evolution of Palladium, a governed execution platform, and Origina, which demonstrates governed identity and memory operating above multiple foundational model -- independent of the provider.
The primary work will be led my me, building on several years of applied research and implementation. The concrete outputs include a publically available architectural spec, reference implementation patterns, technical documents, and open demonstrations showing how deterministic governance can reduce operational AI risk.
Theory of Impact
Updated 07/03/26 · By grantmaking.aiAs AI becomes more capable, the primary challenge is not just model behavior. It is ensuring that AI can be deployed safely, consistently, and can be held accountable in real world environments.
Today, governance is often tightly coupled to individual model providers or implemented as application-specific logic. That makes it difficult to maintain consistent oversight as organizations adopt new models and increasingly autonomous workflows.
GIAS reduces this operational risk by defining an architectural separation between AI reasoning and AI governance. Governance of models cannot happen inside the model. GIAS places deterministic policy enforcement, identity, execution, auditability, and forensic replay outside the model. This allows organizations to apply consistent governance regardless of which foundational model is used.
GIAS governs AI systems without requiring changes to the models themselves.
GIAS enabled organizations to retain meaningful human accountability while benefiting from continued advances in AI capability. I believe reducing systemic AI risk requires a trustworthy architecture, not just increasingly capable models.
People
Updated 07/03/26 · By grantmaking.aiTeam Member
Discussion
No comments yet. Be the first to share your thoughts.