Then, you have the second and third layers regarding built-in censorship triggers and ideological training data. But those relate to the model’s weights. The “hazardous product” designation addresses the first layer.