logo
language iconarrow
menu icon

ISO/IEC 42001:2023

42001-19 Possible objectives and risk sources

3 min read · V, IA

All articles

For: V, IA
Source: ISO/IEC 42001:2023(en), English. Annex C, heading “informative”

C.1 says that the annex outlines possible organizational objectives, risk sources and descriptions that the organization can consider in risk management. It is not intended to be exhaustive or applicable to every organization. The organization should determine the objectives and risk sources that are relevant. ISO/IEC 23894 gives more detailed information on these objectives and sources and on their relationship to risk management. That is not available. Evaluation of AI systems, initially, periodically and when prompted, provides evidence that a system is assessed against organizational objectives. That is a description in the annex, not an additional audit interval.

The objectives that C.2 names are:

  • Accountability, C.2.1. The use of AI can change existing accountability frameworks. Where persons previously answered for their actions, actions can now be supported by an AI system or be based on it.
  • AI expertise, C.2.2. A selection of professionals with interdisciplinary skills is needed.
  • Availability and quality of training and test data, C.2.3. Systems based on machine learning need training, validation and test data in order to be trained and tested for the intended behaviour.
  • Environmental impact, C.2.4. The use of AI can have positive and negative impacts on the environment.
  • Fairness, C.2.5. The inappropriate application of AI systems for automated decision-making can be unfair for particular persons or groups.
  • Maintainability, C.2.6. It concerns the ability to handle changes to the system in order to correct faults or to adapt to new requirements.
  • Privacy, C.2.7. Misuse or disclosure of personal and sensitive data, as an example health records, can have harmful effects for the persons affected.
  • Robustness, C.2.8. Robustness properties show the ability or inability of the system to have a performance on new data comparable to that on the data with which it was trained, or on the data of typical operation.
  • Safety, C.2.9. Safety concerns the expectation that a system, under defined conditions, does not lead to a state in which human life, health, property or the environment is endangered.
  • Security, C.2.10. For AI, and in particular for systems based on machine learning, new security issues should be considered, beyond classical information and system security.
  • Transparency and explainability, C.2.11. Transparency concerns characteristics of the organization that operates AI systems, and of the systems themselves. Explainability concerns explanations of important factors that influence the results, provided to interested parties in a way understandable to humans.

The risk sources that C.3 names are: complexity of the environment, C.3.1; lack of transparency and explainability, C.3.2; degree of automation, C.3.3; risk sources related to machine learning, including data quality and the manner of collection, as an example Data Poisoning, C.3.4; hardware, including faulty components or the transfer of trained models between systems, C.3.5; issues across the life cycle, as an example shortcomings in the design, inadequate deployment, lack of maintenance, issues of decommissioning, C.3.6; technology readiness, both little-mature technology because of unknown factors and mature technology because of carelessness, C.3.7.

6.2 refers to Annex C as a non-exhaustive list of objectives related to risk management. Neither 6.2 nor Annex C requires an objective or a treatment for every one of these names. C.1 says that the organization should determine what is relevant.

Not in this text: a shall-list, an evaluation scale, and the content of ISO/IEC 23894. Anyone who treats a missing name from C.2 or C.3 as a nonconformity is reading an informative annex as a requirement. That contradicts the heading and C.1. That is also covered in 42001-21.