We refer reviewers to the ACM's policies on reviewers .
Disclosing submitted paper contents (even indirectly in the form of a draft of a paper review) to a non-confidential AI tool puts unpublished work in jeopardy and reviewers could be held liable for breaching confidentiality. Furthermore, the expectation for program committee members has always been to destroy copies of paper submissions they have reviewed. This, however, would be challenging to enforce when Generative AI / LLM tools are used.
Reviewers must therefore not use Generative AI / LLM tools to assist in any way whatsoever with their reviews.