I agree SAS - it's down to education, training and practice but since civil type rating is a very quick process compared to converting to a new type in the military and there often aren't the non-revenue hours to give the practice, is it any wonder that pilots don't really know their AP systems?
Having taught on the 365N3 (3-axis) and as a recent convert to the 412 (again 3-axis) I am very comfortable with using the AP but that is after 500 hours of instructing it on the N3.
Although helping the cyclic channel capture the GS by lowering the lever and assisting the go-around by raising it are what would be considered 'normal' helicopter control inputs (ie what tyou would do if you were flying it manually), it is the conflicting element of controlling IAS with collective on approach that I believe causes problems.
It is the 'half' level of automation that is dangerous - it lulls you into a false sense of security because the AP is doing the hard work for you (where you would otherwise be working quite hard, especially in difficult conditions) and there is the problem of under-arousal at a critical stage of flight ie not visual at DA/DH.
Add in fatigue and body clock issues and its just a question of when not if.