In the book "Computer One" by Warwick Collins, an interesting idea is floated (not sure if it has appeared elswhere). It is that if you viewed virii as subject to "natural" selection or an analogous process, the most successful virii should be those that can exist and replicate efficiently whilst being almost undetectable. This would mean virii that did *nothing* except exist and replicate, hid well, used minimum resources, and generally did not impact a host system at all.

This produces the interesting result that the more you accept this is true, and likely to have happened, the less likely it is we would know it. In this novel, he postulates that there are many such entities going undected in systems and mutating/reproducing constantly! An interesting idea I thought.

You could extend this by looking at bacteria that live in the human gut, and say that the most successful virus should be one that makes its host healthier, and provides a useful service (e.g. cleaning registry, fixing MS bugs, optimising networking).

The scenarios of intelligent agents/genetic algorithms running amok have been well covered, but I thought this was an interesting take on the idea that could be relevant to this thread.