There are simultaneously a large number of laws prohibiting employers from retaliating against persons for voting, and a number of accusations of retaliation for voting. So this isn't a theoretical issue. I'm not sure it's distinct from other methods of compromising trusted users -- the effects are similar whether the compromised node was beaten with a wrench, got brain-eaten, or just trusted Microsoft with their Certificates -- but it's a good demonstration that you simply can't trust any node inside a network.

(There's some interesting overlap with MIRI's value stability questions, but they're probably outside the scope of this thread and possibly only metaphor-level.)

Interestingly, there are some security metrics designed with the assumption that some number of their nodes will be compromised, and with some resistance to such attacks. I've not seen this expanded to reputation metrics, though, and there are technical limitations. TOR, for example, can only resist about a third of its nodes being compromised, and possibly fewer than that. Other setups have higher theoretical resistance, but are dependent on central high-value nodes that trade resistance against to vulnerability against spoofing.

It seems like there's some value in closing the gap between carrier wave and signal in reputation systems, rather than a discrete reputation system, but my sketched out implementations become computationally intractable quickly.