The AI Assurance (AIA) Landscape supports exploration and discovery of assurance risks in a way that maximizes use of AI-enhanced systems while minimizing the risks to the sociotechnical systems in which they are deployed.
More than 50 frameworks and over 500 reports on AI assurance have been published in recent years by various U.S. and international government departments or agencies as well as by industry, not-for-profit and non-government organizations (Fjeld et al., 2022; Shneiderman, 2022). Despite the prevalence of AI assurance frameworks, there is a conspicuous absence of a standardized approach to assuring AI-enabled systems and a notable absence of a common AI assurance scheme or vocabulary (cf. Robbins et al., 2024).
The AI Assurance (AIA) Landscape provides a first step towards integrating existing frameworks in a domain-agnostic manner, and identifying a comprehensive list of standardizing assurance needs and definitions that would facilitate collaboration across stakeholders and generalization across technologies, disciplines, and domains.
It synthesizes 50+ existing AI assurance frameworks, including NIST’s AI RMF, and contains a comprehensive set of AIA needs and requirements, including 11 AI assurance categories, 66 specific assurance needs and associated definitions of terms, mapped against existing frameworks.
The AIA Landscape was developed for use in MITRE’s Risk Discovery Protocol for AI Assurance; a process designed to support exploration of assurance needs to maximize the use of AI-enhanced systems while minimizing the risks to the sociotechnical systems in which they are deployed.