"and there’s no way to detect it"I can see how the type of modification described in the article would not be visually detectable, but wouldn't the bias in the rng be trivially detectable by generating a large quantity of random numbers and testing their distribution?
