Deliberate being the key word here.
If I suck at sampling and make a bias model, I suck at sampling. If I shape a model to neuter it and ensure that it only puts out morally approved outputs, that slants significantly heavier towards censorship