I would believe there to be a significant, non-zero risk that in the R&D process of developing a 1st law neural net, a malicious neural net is created first.