I'm not sure how exactly you'd do it. I guess you could write some kind of AI that writes gadgets, then tries to find the optimal probability of correctly leak

I'm not sure how exactly you'd do it. I guess you could write some kind of AI that writes gadgets, then tries to find the optimal probability of correctly leaked data. There are methodologies for doing this automatically published but it's in the "floats rather than books" category of behaviour since it's quite chaotic, so a formal proof would be hard.