512

Position: A taxonomy for reporting and describing AI security incidents

Main:8 Pages
1 Figures
Bibliography:4 Pages
3 Tables
Abstract

AI systems are vulnerable to attacks, and corresponding AI security incidents have been described. Although a collection of safety incidents around AI will become a regulatory requirement, there is no proposal to collect AI security incidents. In this position paper, we argue that a proposal should be made, taking into account the interests and needs of different stakeholders: industry, providers, users, and researchers. We thus attempt to close this gap and propose a taxonomy alongside its requirements like machine readability and link-ability with existing databases. We aim to spark discussions and enable discussion of which information is feasible, necessary, and possible to report and share within and outside organizations using AI.

View on arXiv
Comments on this paper