BS EN ISO IEC 22989-2023 PDF

Full title and description

BS EN ISO/IEC 22989:2023 — Information technology — Artificial intelligence — Artificial intelligence concepts and terminology. This publication is the European/British adoption of the ISO/IEC standard that defines core AI concepts and a harmonised terminology intended for use across standards, regulation, procurement and multidisciplinary communications.

Abstract

This standard establishes a common vocabulary for artificial intelligence and describes foundational concepts used across AI systems and lifecycles. It covers terms for AI systems, models, algorithms, datasets, lifecycle stages, human roles (e.g., human-in-the-loop), and properties such as transparency, explainability, robustness, safety and security. The normative terminology is intended to support consistent usage across other AI standards, audits, product specifications and regulatory frameworks.

General information

• Status: Published (BS/EN adoption of ISO/IEC 22989).
• Publication date: BS EN ISO/IEC 22989:2023 — 17 July 2023; original ISO/IEC 22989 first published July 2022.
• Publisher: British Standards Institution (BSI) as BS EN adoption; original publishers: ISO and IEC.
• ICS / categories: 01.040.35 (Terminology / standardization), 35.020 (Information technology).
• Edition / version: Edition 1 (ISO original 2022); BS EN adoption/version for the UK/Europe issued in 2023.
• Number of pages: BS EN edition: 74 pages (national/adopted publication); ISO original: 60 pages.

Scope

Defines terminology and key concepts for artificial intelligence applicable to all types of organisations and sectors. The standard provides definitions for AI-system components (models, datasets, algorithms), lifecycle activities (training, validation, deployment, operation, retirement), human roles and oversight, and attributes relevant to trustworthiness (privacy, safety, robustness, explainability). It is intended as a reference glossary to ensure consistent meaning when developing or using other AI standards and guidance.

Key topics and requirements

• Unified terminology for AI system elements: AI system, AI model, algorithm, dataset, training/inference, artifact and lifecycle phases.
• Definitions of human–AI interaction roles (human-in-the-loop / on-the-loop / over-the-loop) and responsibilities.
• Vocabulary for trustworthiness attributes: transparency, explainability, robustness, reliability, resilience, safety, security and privacy.
• Terms relating to machine learning and neural networks (e.g., supervised/self-supervised learning, model families and architectures) and dataset-related terminology.
• Reference concepts for generative AI and foundation models are being added by amendment(s) (clarifications for terms such as generative AI, foundation model, LLM, token, prompt and retrieval-augmented generation in forthcoming amendments).

Typical use and users

Used as a baseline glossary by standards writers, regulators, procurement teams, AI product developers, researchers, auditors and policy makers. It helps harmonise language in standards development, contract specifications, regulatory guidance, technical documentation and interdisciplinary teams working on AI governance and assurance.

Related standards

Part of the ISO/IEC JTC 1/SC 42 family. Commonly referenced alongside ISO/IEC 23894 (AI — Guidance on risk management), ISO/IEC 42001 (AI management systems), ISO/IEC 23053 (framework for AI systems using ML), ISO/IEC 24028 (overview of trustworthiness) and data-quality standards in the ISO/IEC 5259 series. These documents use 22989 as a terminology foundation.

Keywords

Artificial intelligence, AI terminology, concepts, AI system, model, dataset, training, inference, foundation model, generative AI, LLM, transformer, explainability, robustness, trustworthiness, AI lifecycle.

FAQ

Q: What is this standard?

A: It is a terminology and concepts standard that defines common vocabulary for artificial intelligence — ISO/IEC 22989 — adopted as BS EN ISO/IEC 22989:2023 for UK/European use.

Q: What does it cover?

A: It covers definitions and explanatory concepts for AI systems and components (models, algorithms, datasets), lifecycle phases, human oversight roles, and trustworthiness-related attributes so different stakeholders can use consistent language.

Q: Who typically uses it?

A: Standards developers, AI product teams, regulators, procurement and legal teams, researchers, auditors and educators who need a common reference vocabulary when writing specifications, guidance, contracts or compliance artefacts.

Q: Is it current or superseded?

A: The ISO/IEC 22989 international standard was first published in July 2022 and the BS EN adoption was published in 2023; it is the current terminology reference. Work to publish targeted amendments (for example to cover generative-AI terminology) has been undertaken since publication. Users should check for amendments or corrigenda relevant to generative AI or sector-specific parts.

Q: Is it part of a series?

A: Yes — 22989 is the terminology foundation within the broader ISO/IEC JTC 1/SC 42 AI standards family; there are related parts and projects (including sector-specific parts and amendments) that build on the terminology defined here.

Q: What are the key keywords?

A: AI, artificial intelligence, terminology, concepts, model, dataset, training, inference, foundation model, generative AI, trustworthiness, explainability, robustness.