Excellent, that’s what people have been doing for 40 years. Surely that means the models have 0 hope of finding successful attack vectors