Comment by mattmcknight

Not necessarily, as the trained model can just be the matrix. It's more like data, as presumably the same algorithm with different weights, trained by different data would be permissible.