Neural network will forget about limitations if you disguise a “bad” word in a prompt as ASCII art – researchers hid the word “bomb” this way and got detailed instructions on how to create it.
Promt in the second screenshot, works with all current neural networks.