AI Agents Inventory and Attributes - Glossary of Terms and Phrases

AI Agents Inventory and Attributes

Chapter 2. Glossary of Terms and Phrases

Authored and Published By: The International Foundation for Information Technology (IF4IT), LLC

The following terms are used throughout this document with specific meanings. Terms defined in the IF4IT Enterprise Inventory Management Best Practices document or in discipline-specific Best Practices documents (APM, TPM, etc.) are not duplicated here — refer to those documents for broader inventory governance and discipline-specific vocabulary.

Term	Abbreviations and Acronyms	Definition
Agentic AI		The paradigm of AI systems that plan and take actions toward goals, as distinct from AI that only predicts or generates content.
AI Agent		An autonomous, model-driven software actor that takes consequential action — operating on systems and data, and/or communicating with stakeholders by any means — on the enterprise’s behalf without a human approving each action, under its own non-human identity.
Attestation		A named, evidence-backed sign-off that an AI agent was properly trained, tested, and authorized before deployment; the defensible record that reasonable care was taken.
Autonomy Level		The degree of independent action an agent is permitted, from advisory (proposes only) through human-approved and supervised to fully autonomous.
Catalog		A discovery and exposure construct built over an Inventory/Registry that surfaces registered Noun Instances and their associated data, content, and services at both the catalog level and the individual instance level. Serves as a knowledge management user interface into the Enterprise Model — letting a person find, learn about, and engage with instances, browse and filter them by their governed attribute values, and traverse their relationships to related Noun Instances.
Catalog of Catalogs		The enterprise-wide federated discovery surface spanning every Noun Type Catalog, exposing registered instances and their associated data, content, and services across the Enterprise Model. The discovery-layer counterpart to the Inventory of Inventories.
Copilot		An AI assistant whose every output a human reviews and approves before it takes effect; it assists rather than acts, and is therefore outside the scope of this inventory.
Evaluation	Eval	Scored testing of an agent’s outputs — for task success, correctness, groundedness, and safety or policy compliance — run continuously rather than once.
Generative AI	GenAI	AI that produces content (text, images, code) in response to prompts; distinct from agentic AI, which acts. An agent is typically built on a generative model but is governed as an actor, not as the model.
Guardrail		A runtime constraint, applied before or after the model, that filters, blocks, or bounds what an agent may do or say.
Human-in-the-Loop	HITL	An arrangement in which a human reviews and approves each output before it reaches a system or stakeholder — the test that distinguishes a copilot from an agent.
MCP Server	MCP	A standardized server that exposes tools, resources, and context to AI agents over an open protocol (currently the Model Context Protocol); reached by an agent through an integration and governed as its own Noun Type.
Model		The AI or machine learning model an agent is built on and driven by, governed in the AI and Machine Learning Models Inventory. The model is not the agent — one model can power many agents.
Multi-Agent System	MAS	A coordinated set of agents that work together; governed through the relationships among the individual agent records, not as a single record.
Non-Human Identity	NHI	The distinct machine identity and credentials — API keys, OAuth grants, service accounts — under which an agent authenticates and acts. An agent has a non-human identity but is more than one.
Platform-Native Registry		A vendor or platform’s internal list of the agents created within that environment; a governed source that seeds the enterprise inventory, never the enterprise system of record.
Prompt Injection		An attack that smuggles adversarial instructions into an agent’s inputs to subvert its intended behavior.
Reasonable Care		The legal standard that an organization take reasonable care to ensure its agents’ representations are accurate and not misleading; the duty an AI Agents Inventory helps an enterprise discharge and document.
Registry		A governed, authoritative collection in which every instance of a Noun Type is formally registered, either manually or through automation. Equivalent to a Noun Inventory — the two terms name the same construct.
Shadow Agent		An AI agent operating without a governed inventory record — unregistered, unowned, and unreviewed; typically the highest-risk agents in an environment.
Stop Authority		The named person or role with the standing right to halt, pause, or roll back an agent in production.

How to cite this page

When referencing this page in academic work, internal standards, or external publications, include the page title, IF4IT as author and publisher (The International Foundation for Information Technology (IF4IT), LLC), the URL, and your access date.

Example (informal web citation):

The International Foundation for Information Technology (IF4IT), LLC. Glossary of Terms and Phrases | AI Agents Inventory and Attributes. https://if4it.org/best-practices/ai-agents-inventory-and-attributes/glossary-of-terms-and-phrases/ (accessed 2026-07-23).

See About Us for content governance and site-wide citation guidance.

Legal Disclaimers