Artificial Intelligence (AI) technologies are increasingly being adopted for use in autonomous systems due to the improved performance they offer. However, despite the growing prevalence of autonomous systems there is limited guidance available on the integration of new AI-based components into existing systems, and with other AI components. To gain a greater understanding of this assurance need, a taxonomy of AI-based autonomous component integration faults has been created. This taxonomy is intended to help guide the user towards appropriate mitigation approaches for each category of fault. This paper presents the current working taxonomy, as well as a discussion of potential further work.