# What the ISO 22989 amendment changes

Published contents
Contents1 Scope2 Normative references3 Terms and definitions4 Abbreviated terms5 AI conceptsClause 3, terms
3.1 Terms related to AI3.1.1 AI agent3.1.2 AI component3.1.3 artificial intelligence3.1.4 artificial intelligence system3.1.5 autonomy3.1.6 application specific integrated circuit3.1.7 automatic3.1.8 cognitive computing3.1.9 continuous learning3.1.10 connectionism3.1.11 data mining3.1.12 declarative knowledge3.1.13 expert system3.1.14 general AI3.1.15 genetic algorithm3.1.16 heteronomy3.1.17 inference3.1.18 internet of things3.1.19 IoT device3.1.20 IoT system3.1.21 knowledge3.1.22 life cycle3.1.23 model3.1.24 narrow AI3.1.25 performance3.1.26 planning3.1.27 prediction3.1.28 procedural knowledge3.1.29 robot3.1.30 robotics3.1.31 semantic computing3.1.32 soft computing3.1.33 symbolic AI3.1.34 subsymbolic AI3.1.35 task3.2 Terms related to data3.2.1 data annotation3.2.2 data quality checking3.2.3 data augmentation3.2.4 data sampling3.2.5 dataset3.2.6 exploratory data analysis3.2.7 ground truth3.2.8 imputation3.2.9 input data3.2.10 label3.2.11 personally identifiable information3.2.12 production data3.2.13 sample3.2.14 test data3.2.15 validation data3.3 Terms related to machine learning3.3.1 Bayesian network3.3.2 decision tree3.3.3 human-machine teaming3.3.4 hyperparameter3.3.5 machine learning3.3.6 machine learning algorithm3.3.7 machine learning model3.3.8 parameter3.3.9 reinforcement learning3.3.10 retraining3.3.11 semi-supervised machine learning3.3.12 supervised machine learning3.3.13 support vector machine3.3.14 trained model3.3.15 training3.3.16 training data3.3.17 unsupervised machine learning3.4 Terms related to neural networks3.4.1 activation function3.4.2 convolutional neural network3.4.3 convolution3.4.4 deep learning3.4.5 exploding gradient3.4.6 feed forward neural network3.4.7 long short-term memory3.4.8 neural network3.4.9 neuron3.4.10 recurrent neural network3.5 Terms related to trustworthiness3.5.1 accountable3.5.2 accountability3.5.3 availability3.5.4 bias3.5.5 control3.5.6 controllability3.5.7 explainability3.5.8 predictability3.5.9 reliability3.5.10 resilience3.5.11 risk3.5.12 robustness3.5.13 stakeholder3.5.14 transparency3.5.15 transparency3.5.16 trustworthiness3.5.17 verification3.5.18 validation3.6 Terms related to natural language processing3.6.1 automatic summarization3.6.2 dialogue management3.6.3 emotion recognition3.6.4 information retrieval3.6.5 machine translation3.6.6 named entity recognition3.6.7 natural language3.6.8 natural language generation3.6.9 natural language processing3.6.10 natural language processing3.6.11 natural language understanding3.6.12 optical character recognition3.6.13 part-of-speech tagging3.6.14 question answering3.6.15 relationship extraction3.6.16 sentiment analysis3.6.17 speech recognition3.6.18 speech synthesis3.7 Terms related to computer vision3.7.1 computer vision3.7.2 face recognition3.7.3 image3.7.4 image recognitionClause 5, concepts
5.1 General5.2 From strong and weak AI to general and narrow AI5.3 Agent5.4 Knowledge5.5 Cognition and cognitive computing5.6 Semantic computing5.7 Soft computing5.8 Genetic algorithms5.9 Symbolic and subsymbolic approaches for AI5.10 Data5.11 Machine learning concepts

Home / The amendment / What changes

# What changes, and what you can do now

The amendment adds generative AI to a vocabulary that was written before generative AI was a common product category. The public record names the subject but not the added entries. What a governance lead can do now is make the existing vocabulary explicit, so the change lands on a register rather than on memory.

## The gap the amendment fills

The 2022 standard has entries for the AI system, the model, machine learning, neural networks, deep learning, natural language generation and natural language understanding. It has no entry for generative AI itself. That absence is not a defect in the 2022 text; it is a record of when the text was written.

The amendment is titled Generative AI. ISO publishes that title, the fact that it amends 22989:2022, and the stage. It does not publish the added terms. Anyone who tells you the exact new definitions before publication is guessing or has a licensed draft.

## The 2022 terms a generative AI system already leans on

These entries are public and stable. A generative AI system description will use most of them, so they are the place to start the register.

Public entries the amendment builds onClauseTermMeaning in our words
3.1.4artificial intelligence systeman engineered system whose outputs are things such as content, forecasts, recommendations or decisions, produced for objectives set by people
3.1.23modela physical, mathematical or other logical representation of a system, entity, phenomenon, process or data
3.2.5dataseta set of data sharing one format
3.2.9input datadata an AI system uses to work out a predicted output or an inference
3.3.5machine learningthe process of tuning the parameters of a model by computational means so that the behaviour reflects data or experience
3.3.7machine learning modela mathematical construct that produces an inference or prediction from input data or information
3.4.4deep learningan approach that builds rich hierarchical representations by training neural networks that have many hidden layers
3.4.8neural networka network of one or more layers of neurons joined by weighted links whose weights can be adjusted, taking input data and producing an output
3.5.7explainabilitya property of an AI system that lets it set out the important factors behind its results in terms people can follow
3.5.16trustworthinessthe ability to meet what stakeholders expect in a way that can be verified
3.6.8natural language generationa task that turns data carrying meaning into natural language
3.6.11natural language understandingthe extraction of information by a functional unit from text or speech given in natural language, plus a description of both the input and what it represents

## Five moves to make before publication

- Put the vocabulary under one owner. A named person, a review date and a version. Without that, the amendment arrives as a reading exercise rather than a change.

- Use the public entries in the artefacts you already have. The AI inventory, the model register, the data schedule and the assurance argument should each cite the clause they rely on.

- Record the local synonyms. Every firm has words that are not in the standard. Put them beside the entry they stand for, or mark them as local terms with their own definition.

- Map to ISO/IEC 42001 clause 3. The management system standard uses the same AI vocabulary. A mapping record shows where the two agree and where the firm has chosen a narrower meaning.

- Diarise the stage, not the rumour. Stage 50.20, 60.00 and 60.60 are public fields. A watch on those fields is more reliable than a newsletter.

## What not to do

Do not rewrite the policy around terms you have not read. Do not buy a kit that claims to contain the amendment text. Do not treat a planning date as an ISO deadline. The date on this site is an estimate and is labelled as one wherever it appears.

## Common questions

Does the amendment change ISO/IEC 42001?
No. It amends ISO/IEC 22989:2022. ISO/IEC 42001 is a separate standard. The vocabulary is shared in practice, so a mapping is useful, but the amendment does not rewrite the management system requirements.
Can we prepare without the text?
Yes, at the level of the register: owners, citations, synonyms, mappings and a review date. The exact new definitions can only be added when ISO publishes them.
Is the 2022 standard still current?
Yes. It was published on 19 July 2022 and remains at stage 60.60. The amendment will change it when published; it does not withdraw it now.

Source. ISO catalogue record 88145, read on 21 October 2026. The public clause 3 entries come from the ISO Online Browsing Platform. Claim level: public record and public definitions only. No amendment text is reproduced.

Last updated: 2026-10-21 · Markdown version

Source: https://iso-22989-revision-checker.theartofservice.com/what-changes/