Architecture

Reference architectures for AI-native systems.

Detailed architectural references for the systems described by the governance framework. Each architecture is independently usable, mapped against the relevant governance section, and grounded in a concrete implementation surface (cloud services, policy languages, retrieval boundaries).

Published

AI IAM Reference Architecture

Published

A layered authorization model for AI execution chains. Governs users, agents, tools, data, models, and outputs through continuous runtime enforcement.

Master Diagram→Layered ReferenceInteractive→Cedar Policy Model→Secure RAG→Audit & Observability→AWS Implementation→

Open AI IAM Reference Architecture

Secure RAG

Published

Retrieval-augmented generation as an authorization boundary, not a search mechanism. Five enforcement phases from pre-retrieval policy through output review, with explicit defense against indirect prompt injection.

Five Enforcement Phases→Indirect Prompt Injection→Decision Function→Governance Mapping→

Open Secure RAG

Agentic Workflows

Published

Multi-agent orchestration with bounded delegation, agent-to-agent authorization, and full execution traceability. Treats every agent as a first-class identity.

Five Principles→Four Authorization Decisions→Six Components→Six Failure Modes→

Open Agentic Workflows

AWS-Native AI Deployment

Published

End-to-end patterns for governed AI on AWS. Bedrock, Knowledge Bases, IAM Identity Center, Verified Permissions with Cedar, and centralized observability, plus the secure defaults and account topology that make it operable.

Seven Layers on AWS→Account Topology→Org-level SCP Defaults→Runtime Flow→

Open AWS-Native AI Deployment

View Governance Framework Whitepapers