Explanations of ISO/IEC 42001:2023
Source is ISO/IEC 42001:2023(en). Title: Information technology — Artificial intelligence — Management system. It was prepared by the joint…
Read articleISO/IEC 42001:2023
For: V, IA
Source: ISO/IEC 42001:2023(en), English. Clause 3. ISO/IEC 22989:2022 applies as well, but is not available.
For this document, the terms from ISO/IEC 22989 and those named in Clause 3 apply. What 22989 understands in detail by AI system, machine learning, life cycle or the roles is not replaced here.
Organization, 3.1, is a person or group with its own functions, responsibilities, authorities and relationships in order to achieve its objectives. Note 2: if the organization is part of a larger entity, the term designates only the part that lies within the scope of the AI management system.
Interested party, 3.2, can influence a decision or activity, be influenced by it, or perceive itself to be influenced by it. An overview of interested parties in relation to AI is, according to the note, in ISO/IEC 22989:2022, 5.19. That is not available.
Top management, 3.3, directs and controls at the highest level. It can delegate authority and provide resources. If the scope covers only a part, it is those who direct and control that part.
Management system, 3.4, is the set of interrelated elements for establishing policies, objectives and processes to achieve the objectives. It can address one or more disciplines.
Policy, 3.5, is intentions and direction, formally expressed by top management. Objective, 3.6, is the result to be achieved. Note 4: in the context of this system the organization sets AI objectives consistent with the AI policy in order to achieve specific results.
Risk, 3.7, is the effect of uncertainty. An effect is a deviation from the expected, positive or negative. Note 5 to 6.1.1 says later that this definition allows a broad view, adaptable to sectors, and that it is the task of the organization, within the framework of the risk assessment, first to adopt a view adapted to its context. That can include definitions from the sector, for example from ISO/IEC Guide 51. The Guide is not available. The note is not an instruction to put another definition in the place of 3.7 without the organization doing so in its assessment.
Process, competence, documented information, performance, continual improvement, effectiveness, requirement, conformity, nonconformity, corrective action, audit, measurement and monitoring follow the common basic structure. Two notes apply specifically here. Performance, Note 3 to 3.11, relates both to results from the use of AI systems and to results of the AI management system. The context decides. Audit evidence and audit criteria are defined in ISO 19011, according to Note 3 to 3.18. That standard is not available here as the subject of these articles.
Control in the risk sense, 3.21, “control”, is a measure that maintains or modifies risk, or both. It includes any process, any policy, any device, any practice or other conditions or actions with this effect. Note 2: controls do not always produce the intended or assumed effect. The source of the definition is ISO 31000:2018, 3.8, modified. ISO 31000 is not retold.
Governing body, 3.22, “governing body”, means the persons who are accountable for the performance and conformity of the organization. Not every organization, especially not a small one, has a governing body separate from top management. Source ISO/IEC 38500:2015, 2.9, with notes added.
Information security, 3.23, is the preservation of confidentiality, integrity and availability of information. Other properties can be involved. Source ISO/IEC 27000:2018, 3.28. The 27000 edition is not available.
AI system impact assessment, 3.24, “AI system impact assessment”:
formal, documented process by which the impacts on individuals, groups of individuals, or both, and societies are identified, evaluated and addressed by an organization developing, providing or using products or services utilizing artificial intelligence
A formal, documented process by which an organization that develops, provides or uses products or services utilizing AI identifies, evaluates and addresses impacts on persons, groups of persons, or both, and on societies. The process requirements are in 6.1.4 and 8.4, the reference controls in A.5. That is 42001-06. The definition uses “addressed”. 6.1.4 describes assessing the potential consequences, documenting them and taking the results into account in the risk assessment. This article does not derive a treatment catalogue of its own from that one word.
Data quality, 3.25, is the characteristic of data of meeting the organization’s data requirements for a specific context. The source is ISO/IEC 5259-1, under preparation at the time of publication. The series is not retold. The requirement to define data quality requirements and to meet them is in A.7.4, where this control is necessary, see 42001-15.
Statement of applicability, 3.26, “statement of applicability”, is the documentation of all necessary controls and of the justification for inclusion or exclusion. The definition refers in parentheses to 3.23. 3.21 defines “control”, 3.23 defines information security. The sentence itself speaks of controls and of inclusion or exclusion. This explanation does not correct the parenthesis. Note 1: organizations may not need all the controls from Annex A, or they go beyond the list. “May”. Note 2 uses “shall”: all risks identified shall be documented in accordance with the requirements of this document. All risks identified and the measures determined for their treatment shall be reflected in the statement of applicability. That is a note to the term with “shall”, not a free pointer. The process is in 6.1.3, see 42001-05.
Not in this text: the list of terms in ISO/IEC 22989:2022, a definition of its own of fairness, robustness or transparency, and a correction of the parenthesis in 3.26. Fairness and the other objectives from Annex C are examples in an informative annex, see 42001-19.