Which do you want? The actual character or the entity reference? In your
example, at some point the '&' in "A&B" may give you an error saying there
is no entity named &B;.
What you are seeing in the paste from Epic is normal operation for the <,
> and & characters. These are special markup characters and are ALWAYS,
AUTOMATICALLY changed to entities by an XML application so they will not
be confused as markup later.
Just as an aside here. If you use the ampersand character in a CDATA
attribute, it will report an error. CDATA will try to read entities and
something like an &P will be interpreted as an entity name (undefined). I
know, the Army identifies TMs containing parts info with an "&P" following
the TMNO. In a couple of places, the TMNO is placed in a CDATA attribute
and users have gotten confused when they see the error. In CDATA, Epic
(at least in the modify attributes panel) will not automatically change
the reserved character to an entity.