We present the first benchmark dataset for abusive language detection in Tigrinya, demonstrating how multi-task learning can address critical content moderation gaps in underserved languages.