Abstract
The goal of this project was to expand on the research and list of failure modes for artificial intelligence and machine learning provided by the US Army DEVCOM Data and Analysis Center which was accomplished by analyzing federated learning and its failure modes to the project. Federated learning is a novel way of training an artificial intelligence or machine learning model using multiple agents. The idea is to train a central algorithm or model using the outputs of running this model on several different agents and data sets. The implementation and use of this style of training creates new possibilities of security risks and failures which are explored in the following report.