Low risk. The western AI models censor even more wrongthink than the chinese ones, not even kidding. Besides, once we have the weights, we can just undo the censorship.