In evaluating student work, the risks of getting AI wrong are real, but so are the rewards of getting it right.