A lot of security writing about agents is speculative. Moltbot already has real incidents, and they map to the exact mistakes most people make.