if you sample enough, nearly any real world classifier will have a nontrivial false positive rate.
Um, source? Using the link below, that's clearly not the case.
https://www.ojjdp.gov/ojstatbb/crime/ucr.asp?table_in=2