ok, well it's difficult then to see how any soft fork can be implemented without an inherent risk of a hard fork. You're basically saying that pre-fork nodes are blind to the new rules, and so one can break the new rules without the pre-fork nodes being aware, that's perfectly sensible.
Yes indeed, that's what I'm saying.
But you're still wrong. The old nodes are not "accepting the new rules", as you said. They cannot be accepting any new rules, because they are not designed to recognize rules that did not exist before that version of the consensus rules were written, Bitcoin does not (yet) have an algorithm that can see into the future! 
Ok, so I see where the confusion comes from - my original post above is probably not worded in the best way. However, this one is clear IMHO:
transactions that would be accepted by old nodes may be invalid according to the new rules.
But you claim to disagree with that exact quote. Can you please explain what you think is wrong about this formulation?