How easy is it to fool A.I.-detection tools?