Glossary
The vocabulary Asenion’s documentation uses, defined in one place.
A
Activity Log
A tamper-evident, immutable audit trail of every significant action taken in Asenion — who did what, when, from which IP address, and what changed. Maintained at two levels: organization-wide (visible to Org Admins) and per-AI-system (visible to project members). → Activity Log
AI Agent
A component that performs one task (possibly subdivided into subtasks) within a use case. Agents orchestrate work such as web scraping, API calls, RAG pipelines, or other agents. In a multi-agent system, typically one agent maps to a single LLM. Can be subjected to qualitative assessments and quantitative tests. → AI System
AI governance platform
Asenion’s category: an AI governance platform, with capabilities across the whole lifecycle of an AI system — from the policy that governs it before it is built to the witnessed record of what it did in production. → What is Asenion?
AI Model
A model powering a specific function within a use case. Includes predictive models (e.g. XGBoost, Logistic Regression) and LLMs (e.g. GPT-4, Gemini-2.0). Can be subjected to qualitative assessments and quantitative tests. → AI System
AI Project / AI System
The top-level unit of governance in Asenion. An AI project is the simplest way to start tracking an AI initiative — it can be subjected to qualitative assessments (policy packs) and quantitative tests (assurance packs) directly. For more complex or high-risk systems that require detailed compliance assessments, an AI project can capture additional information through AI use cases. → AI System
AI Use Case
A discrete job-to-be-done within an AI project. Each use case is implemented using AI models, AI agents, and datasets, and can itself be subjected to qualitative assessments and quantitative tests. → AI System
Asenion
The AI governance platform. Organizations use it to turn the rules that apply to their AI — regulations, standards, and their own policies — into structured controls, running assessments, tested models and agents, and tamper-evident records of what happened at runtime. → What is Asenion?
Assessment
Stage 2 of the lifecycle: the system of record. A hierarchical inventory of AI systems, models, agents, datasets, and vendor components, with policies attached, owners, reviewers, and approvers assigned, assessments run against policy packs, a role-based review workflow, a full audit trail, lifecycle stages, and reports drawn from the same records the team works in. → What is Asenion?
Assessment Pack, Evaluation Pack, Runtime Bundle
The export artifacts of Asenion’s three governance surfaces, used in deeper technical documentation — an Assessment Pack from Assessment, an Evaluation Pack from Testing, and a Runtime Bundle from Runtime Governance. → What is Asenion?
assessment status
The stage an assessment is currently in, progressing from Not Started through In Progress, In Validation, In Review, Waiting for Approval, to a terminal state of Approved, Rejected, Published, or Archived. Distinct from audit status. → Workflow
Assurance Pack
A grouping of quantitative tests run against an AI project, use case, model, agent, or dataset. The quantitative counterpart to a Policy Pack. → Risk-based Quantitative Tests
attested
A technical adjective, used only where the cryptographic sense is unambiguous. It is not a heading, nav label, or feature name; the mechanism is the witness record. → What is Asenion?
audit status
The stage of an independent audit conducted against an approved assessment, tracked separately from assessment status: Not Audited, Under Audit, Audited, or Audit Rejected. → Workflow
Authoring
Stage 1 of the lifecycle: the capability that turns a regulation, standard, or policy into structured, cited, versioned controls, and classifies each control by how it must be governed. → What is Asenion?
C
classification (three-surface)
The classification of each control by how it must be governed: a documented obligation to assess, a measurable property to test, or a live condition an operating agent consults at the moment of action. It is the hinge that makes the four stages read as one system. → What is Asenion?
connection
A named, organization-scoped integration to an external system — a cloud platform (Azure, AWS, GCP), an MLflow tracking server, a notification channel (Slack, Microsoft Teams), or a ServiceNow instance. Configured by Org Admins and used to enable Model Discovery, test evidence import, and alert delivery. → Integrations
consultation / consult
The runtime call an agent or application makes to Guardian before a bounded action. A consultation produces a witness record and never returns a verdict. → What is Asenion?
control
A structured, governed requirement generated from source text, with citations to the clause it implements, versioned with tamper-evident history, and classified by how it must be governed. → What is Asenion?
Control Bundle
A named category of controls within a policy, grouping related requirements under a single section — for example, “Data Governance” or “Transparency”. Each bundle carries a weight that contributes to the overall policy score. The middle tier of the Policy → Control Bundle → control hierarchy. → Control Bundle
D
Dataset
A dataset used by an AI use case, tracked and evaluated within Asenion. Can be subjected to qualitative assessments and quantitative tests. → AI System
Document Hub
A per-AI-system file manager for governance evidence. Stores documents — design specifications, data cards, model cards, audit reports, vendor documentation — that support an AI system’s assessments. Files can be attached directly to individual controls as evidence. → Document Hub
G
glossary
This page: the public face of Asenion’s terminology, defining the platform’s vocabulary in one place.
Guardian
Asenion’s runtime governance layer. Agents and applications consult it at the moment of a bounded action; it surfaces the policies and guidance that apply and produces a witness record. → What is Asenion?
L
lifecycle configuration
A named set of stages, required policies, and completion and score thresholds that defines how an AI system progresses from development through production. Created by Org Admins and assigned to AI systems within the organization. → Lifecycle Management
M
Model Discovery
The feature that scans a connected cloud platform (Azure, AWS, or GCP) and surfaces deployed AI models for review and import into Asenion as AI system records. The primary mechanism for building an AI system inventory from an existing cloud deployment. → Model Discovery
O
Operational Risk Assessment
The policy automatically assigned to every new AI system on creation. It collects the system’s description, sector, jurisdiction, intended use case, and methodology, identifies its current lifecycle stage, and produces policy recommendations from the completed answers. → Policy
organization
An isolated workspace in Asenion with its own AI systems, policies, assessments, members, and settings. A single user account can belong to more than one organization; no data crosses between them. → Multiple Organizations
organization unit
A named sub-structure within an organization used to group AI systems and members by business function. Unit types include Department, Division, Team, Project Group, and Other; units can be nested to any depth. → Multiple Organizations
P
Policy
A framework, standard, regulation, or testing criteria represented in Asenion as a structured set of control bundles and controls. Policies are assigned to AI systems and assessed against; they can be out-of-the-box (covering external regulations and standards) or organization-created. → Policy
Policy Library
The organization’s central collection of governance policies available for assignment to AI systems. Contains out-of-the-box policies covering external regulations and standards, as well as organization-created policies. A separate Assurance Library tab surfaces quantitative test policies. → Policy Library
Policy Pack
A grouping of policies that assessments run against. → Policy Frameworks
Q
Quickstart
The build path’s starting point: from access to your first governed AI use case, self-serve. → Quickstart
R
risk status
A continuous, three-dimension score of an AI system’s governance posture, each dimension rated Low, Medium, High, or Unavailable: residual risk (overall compliance posture from assessment scores), technology risk (data handling, model characteristics, and infrastructure), and alignment risk (how closely the system’s actual behavior matches its stated policy positions). Updated whenever an assessment or control answer changes. → Risk Status Tracking
Runtime Governance
Stage 4 of the lifecycle: governance at the moment of action, delivered by Guardian. It scopes the category word “governance” rather than claiming it. → What is Asenion?
T
Testing
Stage 3 of the lifecycle: evidence before and beyond deployment. Structured adversarial testing for LLMs and agentic systems, bias and fairness metrics across protected classes, and statistical drift detection. → Testing
Trust & Security
The trust path’s starting point: deployment options, data handling, authentication, and architecture at a security-reviewer’s altitude. → Trust & Security
W
witness record
A tamper-evident record of what was consulted and what guidance Guardian returned, cryptographically chained so that tampering is evident and the record can be verified afterward. It is the mechanism behind runtime evidence — a mechanism name, introduced where it is explained, not a nav or URL label. → What is Asenion?