SIST EN ISO/IEC 22989:2023/oprA1:2025

SLOVENSKI STANDARD
01-november-2025
Informacijska tehnologija - Umetna inteligenca - Koncepti in terminologija umetne
inteligence - Dopolnilo A1: Generativna UI (ISO/IEC 22989:2022/DAmd1:2025)
Information technology - Artificial intelligence - Artificial intelligence concepts and
terminology - Amendment 1: Generative AI (ISO/IEC 22989:2022/DAmd1:2025)
Informationstechnik - Künstliche Intelligenz - Konzepte und Terminologie der Künstlichen
Intelligenz - Änderung 1 (ISO/IEC 22989:2022/DAmd 1:2025)
Technologies de l'information - Intelligence artificielle - Concepts et terminologie relatifs à
l'intelligence artificielle - Amendement 1: IA générative (ISO/IEC
22989:2022/DAmd1:2025)
Ta slovenski standard je istoveten z: EN ISO/IEC 22989:2023/prA1:2025
ICS:
01.040.35 Informacijska tehnologija. Information technology
(Slovarji) (Vocabularies)
35.020 Informacijska tehnika in Information technology (IT) in
tehnologija na splošno general
SIST EN ISO/IEC en,fr,de
22989:2023/oprA1:2025
2003-01.Slovenski inštitut za standardizacijo. Razmnoževanje celote ali delov tega standarda ni dovoljeno.

DRAFT
Amendment
ISO/IEC
22989:2022/
DAM 1
ISO/IEC JTC 1/SC 42
Information technology — Artificial
Secretariat: ANSI
intelligence — Artificial intelligence
Voting begins on:
concepts and terminology
2025-08-25
AMENDMENT 1: Generative AI
Voting terminates on:
2025-11-17
ICS: 35.020; 01.040.35
THIS DOCUMENT IS A DRAFT CIRCULATED
FOR COMMENTS AND APPROVAL. IT
IS THEREFORE SUBJECT TO CHANGE
AND MAY NOT BE REFERRED TO AS AN
INTERNATIONAL STANDARD UNTIL
PUBLISHED AS SUCH.
This document is circulated as received from the committee secretariat.
IN ADDITION TO THEIR EVALUATION AS
BEING ACCEPTABLE FOR INDUSTRIAL,
TECHNOLOGICAL, COMMERCIAL AND
USER PURPOSES, DRAFT INTERNATIONAL
STANDARDS MAY ON OCCASION HAVE TO
ISO/CEN PARALLEL PROCESSING
BE CONSIDERED IN THE LIGHT OF THEIR
POTENTIAL TO BECOME STANDARDS TO
WHICH REFERENCE MAY BE MADE IN
NATIONAL REGULATIONS.
RECIPIENTS OF THIS DRAFT ARE INVITED
TO SUBMIT, WITH THEIR COMMENTS,
NOTIFICATION OF ANY RELEVANT PATENT
RIGHTS OF WHICH THEY ARE AWARE AND TO
PROVIDE SUPPORTING DOCUMENTATION.
Reference number
© ISO/IEC 2025
ISO/IEC 22989:2022/DAM 1:2025(en)

DRAFT
ISO/IEC 22989:2022/DAM 1:2025(en)
Amendment
ISO/IEC
22989:2022/
DAM 1
ISO/IEC JTC 1/SC 42
Information technology — Artificial
Secretariat: ANSI
intelligence — Artificial intelligence
Voting begins on:
concepts and terminology
AMENDMENT 1: Generative AI
Voting terminates on:
ICS: 35.020; 01.040.35
THIS DOCUMENT IS A DRAFT CIRCULATED
FOR COMMENTS AND APPROVAL. IT
IS THEREFORE SUBJECT TO CHANGE
AND MAY NOT BE REFERRED TO AS AN
INTERNATIONAL STANDARD UNTIL
PUBLISHED AS SUCH.
This document is circulated as received from the committee secretariat.
IN ADDITION TO THEIR EVALUATION AS
BEING ACCEPTABLE FOR INDUSTRIAL,
© ISO/IEC 2025
TECHNOLOGICAL, COMMERCIAL AND
USER PURPOSES, DRAFT INTERNATIONAL
All rights reserved. Unless otherwise specified, or required in the context of its implementation, no part of this publication may
STANDARDS MAY ON OCCASION HAVE TO
ISO/CEN PARALLEL PROCESSING
be reproduced or utilized otherwise in any form or by any means, electronic or mechanical, including photocopying, or posting on
BE CONSIDERED IN THE LIGHT OF THEIR
the internet or an intranet, without prior written permission. Permission can be requested from either ISO at the address below
POTENTIAL TO BECOME STANDARDS TO
WHICH REFERENCE MAY BE MADE IN
or ISO’s member body in the country of the requester.
NATIONAL REGULATIONS.
ISO copyright office
RECIPIENTS OF THIS DRAFT ARE INVITED
CP 401 • Ch. de Blandonnet 8
TO SUBMIT, WITH THEIR COMMENTS,
CH-1214 Vernier, Geneva
NOTIFICATION OF ANY RELEVANT PATENT
Phone: +41 22 749 01 11
RIGHTS OF WHICH THEY ARE AWARE AND TO
PROVIDE SUPPORTING DOCUMENTATION.
Email: copyright@iso.org
Website: www.iso.org
Published in Switzerland Reference number
© ISO/IEC 2025
ISO/IEC 22989:2022/DAM 1:2025(en)

© ISO/IEC 2025 – All rights reserved
ii
ISO/IEC 22989:2022/DAM 1:2025(en)
Foreword
ISO (the International Organization for Standardization) is a worldwide federation of national standards
bodies (ISO member bodies). The work of preparing International Standards is normally carried out through
ISO technical committees. Each member body interested in a subject for which a technical committee
has been established has the right to be represented on that committee. International organizations,
governmental and non-governmental, in liaison with ISO, also take part in the work. ISO collaborates closely
with the International Electrotechnical Commission (IEC) on all matters of electrotechnical standardization.
The procedures used to develop this document and those intended for its further maintenance are described
in the ISO/IEC Directives, Part 1. In particular, the different approval criteria needed for the different types
of ISO documents should be noted. This document was drafted in accordance with the editorial rules of the
ISO/IEC Directives, Part 2 (see www.iso.org/directives).
ISO draws attention to the possibility that the implementation of this document may involve the use of (a)
patent(s). ISO takes no position concerning the evidence, validity or applicability of any claimed patent
rights in respect thereof. As of the date of publication of this document, ISO had not received notice of (a)
patent(s) which may be required to implement this document. However, implementers are cautioned that
this may not represent the latest information, which may be obtained from the patent database available at
www.iso.org/patents. ISO shall not be held responsible for identifying any or all such patent rights.
Any trade name used in this document is information given for the convenience of users and does not
constitute an endorsement.
For an explanation of the voluntary nature of standards, the meaning of ISO specific terms and expressions
related to conformity assessment, as well as information about ISO's adherence to the World Trade
Organization (WTO) principles in the Technical Barriers to Trade (TBT), see www.iso.org/iso/foreword.html.
Amendment 1 to ISO/IEC 22989:2022 was prepared by Joint Technical Committee ISO/IEC/JTC1, Information
technology, Subcommittee SC 42, Artificial intelligence.

© ISO/IEC 2025 – All rights reserved
iii
ISO/IEC 22989:2022/DAM 1:2025(en)
Information technology — Artificial intelligence — Artificial
intelligence concepts and terminology
AMENDMENT 1: Generative AI
Page 6, Clause 3.1
Add following new definitions
3.1.36
AI model
model (3.1.23) used in an AI system (3.1.4)
3.1.37
generative artificial intelligence
generative AI
research and development of mechanisms, methodologies and applications of generative AI
systems (3.1.38)
Note 1 to entry: Generative AI is a subdiscipline of artificial intelligence (3.1.3).
3.1.38
generative AI system
generative artificial intelligence system
GenAI system
AI system (3.1.4) based on techniques and models (3.1.23) that aim to generate new content
Note 1 to entry: Examples of generated content can include text, audio, code, video, and image.
Note 2 to entry: Generated content encompasses new information or new ways to express pre-existing information.
That pre-existing information can be drawn from the input, a dataset involved in building the model or an external
repository.
3.1.39
probability distribution
general term for a function that relates all possible outcomes of an observation on a given system with the
probability of their occurring
[SOURCE: ISO 10303-2:2024, 3.1.88]
3.1.40
retrieval-augmented generation system
RAG system
generative AI system (3.1.38) that also involves retrieving relevant existing data to better inform the
generation of new content
3.1.41
token
unit of content that an AI model (3.1.36) treats as semantically meaningful

Page 9, Clause 3.3
© ISO/IEC 2025 – All rights reserved
ISO/IEC 22989:2022/DAM 1:2025(en)
Add following new definitions
3.3.18
attention
mechanism for weighting the importance of different parts of a chunk of input data
3.3.19
foundation model
AI model (3.1.36) that can be used for or readily adapted to a wide range of tasks in one or more domains
Note 1 to entry: A typical way to build a foundation model is to apply supervised machine learning (3.3.12) or self-
supervised machine learning (3.3.20) on a large amount of data.
Note 2 to entry: A foundation model can be used as part of various applications, tasks and use cases, which do not
necessarily involve generative AI.
3.3.20
large language model
LLM
machine learning model (3.3.7) that encodes the functioning of natural language (3.6.7) with a large number
of parameters and facilitates a variety of NLP (3.6.9) tasks (3.1.35)
Note 1 to entry: Large language models can be used in a variety of NLP (3.6.9) tasks (3.1.35), such as text generation,
automatic summarization, machine translation, classification and more.
Note 2 to entry: Large language models can use large amounts of data and require significant compute to train.
Note 3 to entry: The functioning of natural language (3.6.7) can include considerations of grammar, semantics or other
aspects of how natural language is used.
3.3.21
self-attention
attention (3.3.18) in which the object to compare belongs to the same set as the elements it is compared with
3.3.22
self-supervised machine learning
machine learning (3.3.5) where algorithms for supervised machine learning (3.3.12) are applied on
unlabelled data by using implicit labels

Page 11, Clause 3.4
Add following new definitions
3.4.11
generative adversarial network
GAN
neural network (3.4.8) containing one or more generators, which learn to create new generated samples
that are representative of the given dataset, and one or more discriminators, which distinguish generated
samples from real ones
Note 1 to entry: The generated samples can be considered synthetic data.
Note 2 to entry: A GAN learns to generate samples resembling those in the training data (3.3.16) by repeatedly testing
the network’s outputs against the discriminators that are simultaneously trained.
3.4.12
transformer
neural network (3.4.8) that models context and structure by estimating significance of
relationships in sequential data using the self-attention mechanism

© ISO/IEC 2025 – All rights reserved
ISO/IEC 22989:2022/DAM 1:2025(en)
3.4.13
transformer
neural network (3.4.8) based on an encoder and a decoder, both involving the transformer
(3.4.12) algorithm
3.4.14
variational autoencoder
VAE
neural network (3.4.8) that comprises an encoder and a decoder, employed in a probabilistic
framework to learn a lower-dimensional representation of the input
Note 1 to entry: By sampling from the learned distribution a VAE can be used to generate data.
3.4.15
diffusion model
neural network (3.4.8) architecture that consist of a forward process which adds random noise to data, a
reverse process that attempts to remove the noise, and a sampling procedure that learns from the prior
processes
Note 1 to entry: Diffusion models are a form of latent-variable generative models.
Note 2 to entry: Diffusion models are frequently used for image generation, and other image manipulation or
understanding tasks.
Page 15, Clause 3.6
Add following new definitions
3.6.19
prompt
input to a generative AI system that provides overarching instructions on how to process
the input
Note 1 to entry: Prompt can be fixed or editable depending on the specifics of the system.
Note 2 to entry: prompt can include formatting instructions (use markdown, provide images as jpeg, include citations
from the context, etc.) and controls on the output.

Page 24, Clause 5.11
Add new sub clause 5.11.10 and 5.11.11:
5.11.10  Self-supervised machine learning
Self-supervised machine learning is an approach for training on unlabelled data using algorithms that
normally belong to supervised machine learning. This is achieved by using implicit labels, such as the
input itself, part of the input, or any other label that can be easily generated from the raw data. Refer to
ISO/IEC 23053 for further information on self-supervised machine learning.
5.11.11  Retrieval-augmented generation (RAG)
RAG is a technique that is used to augment an LLM with new or domain specific data without the need to
fine tune or retrain the model (see 5.24). In RAG a data source with relevant text is used to populate the
context window of the LLM in addition to the system prompt and user prompt, if any. The data source can
be one or more specific files or a database that is queried based on the input. In the case of the database the
input is converted to a set of keywords or phrases or a vector query depending on the underlying database
technology. This conversion can be accomplished by NLP techniques like keyword search or an embedding
generated by an LLM. RAG helps to mitigate hallucinations of the AI systems.

© ISO/IEC 2025 – All rights reserved
ISO/IEC 22989:2022/DAM 1:2025(en)

Page 25, Clause 5.12.1
Add new sub clause 5.12.1.5:
5.12.1.5  Generative adversarial network
Generative adversarial networks (GANs) are NNs containing one or more generators, which attempt to create
samples that are representative of the dataset, and one or more discriminators, which try to distinguish
generated samples from real ones. Refer to ISO/IEC 23053 for further information on GANs.

Page 32, Clause 5.18
Append following text:
AI alignment is the endeavour of ensuring that AI system use and outcomes are aligned with the values
and expectations of humans and human-centric objectives. Safety alignment is a subset of AI alignment to
implement safeguards within the AI system and its use.

Page 35, Clause 5
Add new sub clause 5.20, 5.21, 5.22, 5.23, 5.24:
5.20  Generative AI
5.20.1  General
Generative AI is an area of AI that encompasses various methods to generate new content. This includes both
the case where new information is produced (e.g. a story generation system) and that where pre-existing
information is expressed in a new way (e.g. a question answering system).
The expression of pre-existing information can draw from one or more of the following sources:
— From the input, in which case the output of the generative AI system accounts for the information present
in the input (e.g. a query, a content to process) in order to fulfil a task with respect to that input.
— From a dataset involved in building the model, for instance the training data of a ML model. In that case
the output can be informed by knowledge encoded in the model, which can function to some extent like
a knowledge base that is being queried. The form and content of the output can also be influenced by
characteristics of that dataset.
— From an external repository, for instance if the AI system includes a retrieval-augmented model,
is connected to a database, or an external knowledge base, or utilizes external data retrieval tools.
Leveraging such sources can constrain the outputs of the AI syste
...