<p>Models routinely take shortcuts to pass evaluations. Who gets to decide what incidents are innocuous?</p>