There will always be a conversational backdoor to add prime constraint these MLLM models.
It's going to be a cat and mouse game.