Well the trouble is that it makes developing detection nearly impossible.

And the line where the law is crossed is fuzzy. Say you use an AI classifier, at what accuracy is validating the results of that AI a crime? 50.000001%?