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. A.9, B.9
The control objective A.9 is to ensure that the organization uses AI systems responsibly and in accordance with the policies of the organization. Three reference controls, to the extent necessary. They apply to use. The development side is in A.6, see 42001-14.
A.9.2: define and document the processes for the responsible use of AI systems.
A.9.3: determine and document objectives that guide responsible use.
A.9.4: ensure that the AI system is used in accordance with its intended uses and the accompanying documentation.
B.9 uses “should”. B.9.2: whether the system is developed by the organization itself or obtained from a third party, the organization should make clear which considerations determine the use, and develop policies on them. Examples: approvals required, costs including ongoing monitoring and maintenance, approved procurement requirements, applicable legal requirements. Existing policies for other systems can be included if that is wanted.
B.9.3 names objectives that an organization can have depending on the context: fairness, accountability, transparency, explainability, reliability, safety, robustness and redundancy, privacy and security, accessibility. “Some objectives include”. That is not a mandatory list. Annex C is the informative catalogue, see 42001-19. Once defined, the organization should implement mechanisms in order to achieve the objectives. It should determine in which life cycle stages objectives of meaningful human oversight should be included. That can comprise: human reviewers with the authority to override decisions of the system; oversight where instructions or other documentation require it for an acceptable use; monitoring of performance, including the accuracy of the outputs; reporting of concerns about outputs and about changes in performance; considering whether automated decision-making is appropriate for a responsible approach and the intended use. The need for oversight can be informed by the impact assessment. Personnel should be informed and trained and should understand the instructions and their tasks. “Should” and “can”. There is no row “human oversight” in Table A.1.
B.9.4: the system should be deployed in accordance with the instructions and the other documentation. Data to which it is applied can have to conform to the documentation in order for the performance to hold. Operation should be monitored. If correct deployment according to the instructions gives rise to concerns because of impacts or legal requirements, the organization should communicate that internally and to third-party suppliers of the system. Event logs or other documentation of the deployment and the operation should be retained to the extent they can show that the system is used as intended, or support the communication of concerns. The duration depends on the intended use, the retention rules and relevant legal requirements.
Not in this text: a guideline on use as a template, an obligation for every example objective in B.9.3, and an obligation, read out of the guidance, to have every automated decision confirmed by a human. A.9.4 requires, to the extent necessary, that use in accordance with the intended use and the accompanying documentation is ensured.