Comment by foo12bar

Why not have a second model that reads every response looking for something nefarious?