ICDM Guidelines for LLM Use in Peer Review

Permitted use cases:

• Reviewers may use generative AI to edit reviewer-written text for grammar, clarity, organization, or conciseness.

• Reviewers may use generative AI for literature search related to a submission, provided they do not upload unpublished manuscript content. Reviewers remain responsible for verifying all outputs.

• Reviewers may consult privacy-preserving LLM tools to support their reading of a submission, provided the tools do not use submitted content for training and have clear data-retention limits. Suitable tools may include enterprise or institutional services, consumer services with explicit no-training settings, and self-hosted models. Reviewers are responsible for writing their own reviews.

Prohibited use cases:

• Reviewers must not use LLMs to generate substantive review content, including paper summaries, strengths, weaknesses, novelty assessments, ethics assessments, score justifications, confidence ratings, or decision recommendations from a submitted manuscript.

• Reviewers must not share unpublished manuscript content, reviews, or discussion content with non-privacy-preserving generative AI systems.

Disclosure of all generative AI use in the review form:

• Any use of generative AI during the review process must be disclosed in the review form.

• Undisclosed or improper AI use may be investigated and may result in sanctions.