Glossary¶
AI system¶
A machine-based system that infers from inputs how to generate outputs such as predictions, content, recommendations, or decisions that can influence physical or virtual environments. Validate terminology against applicable law and standards.
Assurance case¶
A structured, evidence-supported argument that a system is sufficiently trustworthy for a defined use and context.
Consequential action¶
An action that can materially affect rights, access, safety, liberty, finances, basic services, employment, or interests of comparable consequence.
Context envelope¶
The authorized, task-specific set of identity, purpose, data, policy, model, tool, and evidence attributes that governs an AI invocation.
Enterprise Context Architecture (ECA)¶
A cross-cutting architectural perspective that connects business intent to information, systems, and intelligent outcomes. ECA organizes enterprise context into seven domains: People & Org, Business, Information, Technology, Governance, Integration, and Operational. Requirements Management sits at the center, and ECA applies across all TOGAF ADM phases.
Requirements Management¶
The center of the TOGAF ADM and ECA. Requirements Management captures and maintains context requirements so every architecture phase and AI capability draws on the same authoritative context.
TOGAF ADM¶
The Architecture Development Method of the TOGAF standard: an iterative cycle of phases (A through H) for developing and governing enterprise architecture. ECA provides a cross-cutting perspective across all ADM phases.
Grounding¶
Connecting generated output to authoritative or controlled evidence. Grounding reduces error, and systems still require validation.
Human oversight¶
A designed capability for competent people with adequate authority, information, time, and mechanisms to supervise, intervene, override, or stop a system.
Material change¶
A change capable of affecting intended use, risk, obligations, performance, affected populations, security, or assurance conclusions.
Model gateway¶
A controlled mediation service for model access, routing, policy enforcement, telemetry, and provider isolation.
Retrieval-augmented generation (RAG)¶
A pattern that retrieves external information and supplies selected context to a generative model. Systems must validate correctness and source authority separately.
Risk owner¶
The person accountable for managing a risk and escalating acceptance decisions to the appropriate authority.
System boundary¶
The components, actors, data, environments, suppliers, processes, and dependencies included in an assessment or authorization scope.
Trust boundary¶
A point where identity, authority, data, execution, or assurance assumptions change. Teams must apply or re-evaluate controls at that point.