RULER introduces new metrics for verifying machine unlearning at the representation level, addressing limitations of current output-level evaluations. The findings reveal that several unlearning methods still retain traces of forgotten data, highlighting the need for improved verification techniques in AI models.