I’ve talked to multiple people about using LLMs in reviews. If the journal is going to systematically run an LLM to provide feedback on the paper, than that LLM output needs to be available to be run before submission. Why not give authors a chance to fix it before submitting?