On the other hand, the question "compared to what" is important.

Yes. When I read this part of your post:

They can easily produce self-contradictory statements and actions and pursue hidden goals

My reaction was, "so, like humans, then?". 100% of the organic intelligent agents we're familiar with are already not 100% trustworthy, why would we assume that artificial ones ought to be more trustworthy or more easily "aligned"? (Personally I kind of worry that the approach the "alignment" people seem to be taking may result in more psychotic-style behaviors from any AIs that result. If you squeeze something too tight you may get something unpleasant oozing through your fingers.)