That's one problem with correlations. They help to identify general patterns, but exceptions are tolerable in their model. Still, there are a lot of examples (with exceptions) of brightly colored species being either toxic or resistant to the toxins of noxious species.