Glossary

Defined terms used across the AIGS standards and the Framework. Where a term is defined in a standard, that definition governs.

Agent (AI agent): An AI system that can take actions, such as calling tools, reading data or sending messages, with some degree of autonomy.
Agent chain: Two or more agents connected so that the output of one becomes the input of another.
AI system: Any model, copilot, agent, automated flow or service that uses artificial intelligence to produce outputs or take actions for the organization.
AI system inventory: The organization's register of AI systems, with owner, purpose, data classes and status. Required by AIGS 100.
Attribution basis: What a unit of AI consumption is attributed to: the AI system, the owner only, or nothing (AIGS 200).
Basis for conclusions: The section of a standard that explains the Board's reasoning, including alternatives it considered and rejected.
Capability profile: A record of what an agent can take in, read, write and send, the tools it can call, and how autonomously it acts (AIGS 500).
Classification coverage: The share of AI-reachable data sources that have a sensitivity classification (AIGS 400).
Collection freshness: How recently consumption or usage data was successfully collected, as distinct from when usage last occurred.
Comment letter: A written response to an exposure draft, submitted during the comment period and considered in redeliberation.
Complete exposure path: An agent or agent chain holding all four exposure legs.
Consumption event: A recorded unit of AI usage, such as tokens or credits, reported by a provider or platform.
Coverage statement: A statement, attached to any report, of which surfaces or sources were measured and which were not.
Evidence strength: How strongly a claim is known to be true: Declared, Configured, Observed or Verified (AIGS 600).
Exposure draft: A proposed standard published for public comment. It has no authority until issued by the Board.
Exposure leg: One of four capabilities that together make an agent exploitable: untrusted input, private data read, shared write, external send.
False green: A control or report shown as healthy when it is only configured, or when data collection failed. AIGS 600 prohibits it.
Human approval: A required human decision before an agent completes an action with material impact (AIGS 500.7).
Leaver reassignment: Transferring ownership of AI systems held by a departing person to a new owner of record (AIGS 100).
Maturity level: One of five stages in the Framework: Initial, Inventoried, Owned, Measured, Assured. Owned is the baseline.
Non-human identity: A service principal, API key, token or other credential used by software, including agents, rather than a person.
Not measured: The required label for a measure or surface where no data was collected. It must never be shown as zero or clean.
Owner of record: The named, current individual accountable for an AI system (AIGS 100).
Owner unknown: The required state for an AI system whose owner cannot be identified or has left. It is a finding, not a default.
Permitted-use register: A record of which data classes each AI system may access and for what purpose (AIGS 400).
Prompt injection: Instructions hidden in content an AI system reads, intended to make it act against its operator.
Redeliberation: The Board's public reconsideration of a draft in light of comment letters.
Sanctioned AI: An AI service, model or tool approved by the organization for stated purposes and data classes (AIGS 300).
Technical Working Group (TWG): A volunteer group of members that drafts and maintains one standard.
Tool server: A service that exposes tools for agents to call, such as a Model Context Protocol (MCP) server (AIGS 500.6).
Triage order: A ranking used to decide what to review first. It is not a probability of wrongdoing (AIGS 300.6).
Unaccountable spend: AI consumption that cannot be linked to an AI system with an owner of record (AIGS 200).
Unmeasured action: A tool action that has not been classified. It must never be assumed to be read-only (AIGS 500.3).