I get it, but still. Humans are greedy. It wouldn't take many leaks - e.g. an employee in the know instructing a couple of ph.d. friends - to start watering down the performance.
But I guess the track record is a testament that this has not happened to a significant extent. By itself impressive.
The book may shed some light on it, but I’d assume each of their researchers is responsible for and maintains a few algos/strategies, which run on an advanced platform/code maintained by developers. So you’d have to steal their code and then setup a larger team of capable yet corrupt scientists to maintain stolen code and thousands of stolen algos.