Completely impossible to completely filter LLM outputs like this. As hard as all these big companies making frontier models have tried, you can still get them to say and tell you whatever you want if you prompt them right.

Also this was on a fucking Tesla I guess, so its not like this was in anyway a model designed for use with children. In fact, it sounds like it was just the standard Grok model, which is intended to be edgy and sexual. So the model is actually functioning as intended lol