Security and privacy in artificial intelligence use cases — Best practices
Introduction: Artificial intelligence (AI) and machine learning (ML) are increasingly being adopted by the digital industry, using algorithms to make decisions that have the potential to negatively impact the privacy of individuals and in some cases can even cause harm to some of them, unless adequate safeguards are deployed. Such safeguards to protect privacy often depend on a variety of factors including the specific type of process, sensitivity of data used, and potential harm likely to be caused. This concern has been expressed by: — Practitioners, who identified 23 principles for AI at the 2017 Asilomar conference[1] covering research, ethics and values, as well as longer term issues. — Standard developers, as evidenced by the report on ethically aligned design published by the IEEE Global Initiative on Ethics of Autonomous and Intelligent Systems[2]. — Policy makers, as exemplified by the appointment by the European Commission of a high-level expert group on artificial intelligence and the subsequent publication of an assessment list[3]. This document provides an analysis of security and privacy of use cases provided in ISO/IEC TR 24030, which should be used in parallel. A number of additional use cases are provided in Annex A. This document also uses concepts from ISO/IEC TR 24028, which addresses trustworthiness in AI systems, including approaches to establish trust (e.g. transparency, explainability, controllability), and to achieve trustworthiness properties (e.g. resiliency, rel
Scope: This document outlines best practices on assessing security and privacy in artificial intelligence use cases, covering in particular those published in ISO/IEC TR 24030. The following aspects are addressed: — an overall assessment of security and privacy on the AI system of interest; — security and privacy concerns; — security and privacy risks; — security and privacy controls; — security and privacy assurance; and — security and privacy plans. Security and privacy are treated separately as the analysis of security and the analysis of privacy can differ.