ISO/PAS 8800:2024
(Main)Road vehicles — Safety and artificial intelligence
Road vehicles — Safety and artificial intelligence
This document applies to safety-related systems that include one or more electrical and/or electronic (E/E) systems that use AI technology and that is installed in series production road vehicles, excluding mopeds. It does not address unique E/E systems in special vehicles, such as E/E systems designed for drivers with disabilities. This document addresses the risk of undesired safety-related behaviour at the vehicle level due to output insufficiencies, systematic errors and random hardware errors of AI elements within the vehicle. This includes interactions with AI elements that are not part of the vehicle itself but that can have a direct or indirect impact on vehicle safety. EXAMPLE 1 Examples of AI elements within the vehicle include the trained AI model and AI system. EXAMPLE 2 Direct impact on safety can be due to object detection by elements external to the vehicle. EXAMPLE 3 Indirect impact on safety can be due to field monitoring by elements external to the vehicle. The development of AI elements that are not part of the vehicle is not within the scope of this document. These elements can conform to domain-specific safety guidance. This document can be used as a reference where such domain-specific guidance does not exist. This document describes safety-related properties of AI systems that can be used to construct a convincing safety assurance claim for the absence of unreasonable risk. This document does not provide specific guidelines for software tools that use AI methods. This document focuses primarily on a subclass of AI methods defined as machine learning (ML). Although it covers the principles of established and well-understood classes of ML, it does not focus on the details of any specific AI methods e.g. deep neural networks.
Véhicules routiers — Sécurité et intelligence artificielle
General Information
Standards Content (Sample)
Publicly
Available
Specification
ISO/PAS 8800
First edition
Road vehicles — Safety and artificial
2024-12
intelligence
Véhicules routiers — Sécurité et intelligence artificielle
Reference number
© ISO 2024
All rights reserved. Unless otherwise specified, or required in the context of its implementation, no part of this publication may
be reproduced or utilized otherwise in any form or by any means, electronic or mechanical, including photocopying, or posting on
the internet or an intranet, without prior written permission. Permission can be requested from either ISO at the address below
or ISO’s member body in the country of the requester.
ISO copyright office
CP 401 • Ch. de Blandonnet 8
CH-1214 Vernier, Geneva
Phone: +41 22 749 01 11
Email: copyright@iso.org
Website: www.iso.org
Published in Switzerland
ii
Contents Page
Foreword .vi
Introduction .vii
1 Scope . 1
2 Normative references . 1
3 Terms and definitions . 2
3.1 General AI-related definitions .2
3.2 Data-related definitions .7
3.3 General safety-related definitions .9
3.4 Safety: Root cause-, error-and failure-related definitions .11
3.5 Miscellaneous definitions . 12
4 Abbreviated terms . 14
5 Requirements for conformity .15
5.1 Purpose . 15
5.2 General requirements . 15
6 AI within the context of road vehicles system safety engineering and basic concepts .16
6.1 Application of the ISO 26262 series for the development of AI systems .16
6.2 Interactions with encompassing system-level safety activities .17
6.3 Mapping of abstraction layers between the ISO 26262 series, ISO/IEC 22989 and this
document . 20
6.4 Example architecture for an AI system . 22
6.5 Types of AI models . 23
6.6 AI technologies of a ML model . 23
6.7 Error concepts, fault models and causal models .24
6.7.1 Cause-and-effect chain . .24
6.7.2 Root cause classes . 26
6.7.3 Error classification based on the safety impact .27
7 AI safety management . .28
7.1 Objectives . 28
7.2 Prerequisites and supporting information . 28
7.3 General requirements . 28
7.4 Reference AI safety life cycle .31
7.5 Iterative development paradigms for AI systems . 33
7.6 Work products . 34
8 Assurance arguments for AI systems .35
8.1 Objectives . 35
8.2 Prerequisites and supporting information . 35
8.3 General requirements . 36
8.4 AI system-specific considerations in assurance arguments . 36
8.5 Structuring assurance arguments for AI systems .37
8.5.1 Context of the assurance argument.37
8.5.2 Categories of evidence . 38
8.6 The role of quantitative targets and qualitative arguments . 39
8.7 Evaluation of the assurance argument . 40
8.8 Work products .41
9 Derivation of AI safety requirements . 41
9.1 Objectives .41
9.2 Prerequisites and supporting information .42
9.3 General requirements .42
9.4 General workflow for deriving safety requirements .43
9.5 Deriving AI safety requirements on supervised machine learning . 46
9.5.1 The need for refined AI safety requirements . 46
iii
9.5.2 Derivation of refined AI safety requirements to manage uncertainty .47
9.5.3 Refinement of the input space definition for AI safety lifecycle . 50
9.5.4 Restricting the occurrence of AI output insufficiencies . 50
9.5.5 Metrics, measurements and threshold design . 54
9.5.6 Considerations for deriving safety requirements . 55
9.6 Work products . 56
10 Selection of AI technologies, architectural and development measures .56
10.1 Objectives . 56
10.2 Prerequisites . 56
10.3 General requirements . 56
10.4 Architecture and development process design or refinement .57
10.5 Examples of architectural and development measures for AI systems . 58
10.6 Work products .62
11 Data-related considerations .62
11.1 Objectives .62
11.2 Prerequisites and supporting information .62
11.3 General requirements .62
11.4 Dataset life cycle . 63
11.4.1 Datasets and the AI safety lifecycle . 63
11.4.2 Reference dataset lifecycle . 64
11.4.3 Dataset safety analysis . 65
11.4.4 Dataset requirements development .71
11.4.5 Dataset design .74
11.4.6 Dataset implementation .
...
Questions, Comments and Discussion
Ask us and Technical Secretary will try to provide an answer. You can facilitate discussion about the standard in here.