The Pentagon's Joint Artificial Intelligence Center explores ways

that adversaries can exploit AI vulnerabilities

By infiltrating the training data collection and labeling process, enemies can "poison" the model and exploit its consequent weakness

Click here for related article