I've tried some abliterated models. So far, they're not evil - you can make them say evil things, but they don't leap right to it without a bit of pushing. Or perhaps I'm not asking the right questions...