BBC tech journalist Thomas Germain’s simple—and hilarious—experiment exposes a serious flaw in common artificial intelligence tools