ML Systems for Identifying Harmful Content

Imagine you're employed by a social media company. With strict content compliance regulations in place, how would you create a machine learning system to identify harmful content effectively?