This is true for meat-based evolution. In a software-based world, it can be trivial to prove what your source code is, and sometimes it's advantageous too.

For instance, software might be distributed with its code, advertised as doing something. The software benefits if it can convince you that it will do that something ("cooperate"), because in exchange, you run it.

An em that's good at some kind of work might want to convince you to run copies of it: it gets to live, you get useful work done. But you're afraid that it will spy on you or sabotage its work. In a world of ems, the assumption is that nobody knows how to write superhuman (superintelligent) AIs, so we emulate human brains. That means you can't just inspect an em's software to determine what it's going to do, because you don't really understand how it works.

It would be nice if the em could provide a formal proof that in a certain scenario it would behave in a certain way (cooperate), but for software of an em's complexity that's probably never going to be possible.