It was a frontier model from a few months ago. It hasn't done it recently. But it has happened. I imagine if there isn't a lot of training data for an idea it probably has a higher change of generating verbatim, so as time goes on, it should become less of a problem (at least with high quality models).