I recently came across this preprint written by Berner, Herr and Jorgensen (https://arxiv.org/pdf/2611.03358). It is slightly outside my usual areas of interest, though having been written by Palle I still decided to peruse it. Now, I want to clarify that my stance towards the use of LLMs in math has not changed - that is to say, I remain opposed to it. However, if LLMs are to be used, I believe that this preprint has achieved the gold standard of transparency.

It has a rather lengthy four-paragraph AI disclaimer explaining in detail how AI was used and for what purposes; then, for each theorem, they are either unlabeled (indicating that AI was not used), or a summary was given of how AI was used (e.g., it gave an outline of a proof that was later totally rewritten by humans). If LLMs are to be used, I would rather their use be as transparent as possible, as in that preprint.