Can you reverse engineer our neural network?