Page 342 - CyberSecurityEssentialsEnglish

● Remember: Section summary
• Key idea: Integrity concerns the truthfulness of the record, responsibility, and
attribution, not the method of writing alone.
• Practical decision: Agree the permitted uses early, document them, and disclose them
in proportion to their effect and to institutional policy.
• Mistake to avoid: Treating the tool as an author, or using disclosure in place of

verification and responsibility.

7.11 AI in Peer Review and Critical Appraisal
◆ Key idea: confidentiality and judgment do not transfer to a model.

AI can help a researcher examine the logic of their own manuscript or organize review
notes, but peer review involves confidentiality, expertise, judgment, and responsibility. A
confidential manuscript must not be uploaded to a public tool, and the decision to accept
or reject must not be delegated to a model.

Peer review is not an automated test of language or formatting. It is an appraisal of the
question, the method, the analysis, the interpretation, the originality, and the transparency.
AI may help build a checklist, summarize the structure of an article, detect a contradiction

between a table and a paragraph, or phrase a comment more clearly. But it may fail to
understand a disciplinary tradition, a novel method, or a limitation that does not appear in
the abstract.

Confidentiality is the first constraint. A manuscript sent to a reviewer is not theirs to transfer

to an external service. Editorial guidance requires you to keep review materials confidential,
and journal instructions should state clearly how AI may be used [16]. A reviewer therefore
reviews the journal’s policy before any use, and the use may be prohibited even if author

names have been removed.

An author may use a tool as a trial reviewer on their own version, provided data and
ownership are protected. It helps to ask specific questions: does the question match the

design? Where is there a jump from association to causation? Are the table’s results
consistent with the text? What information would be needed to re-run the analysis? These
questions are better than “review the paper”, because they define a criterion.

Good review distinguishes a major issue from a minor one. A major issue changes the validity

of the question or the finding, such as a design that cannot answer the question, an unclear
sample, or an unsuitable analysis. A minor point may concern presentation, a definition, or
a figure. A tool can organize the comments, but ranking them requires understanding the

impact of each problem.

Comments must be actionable and respectful. It is not enough to say “the method is weak”.
Identify the location, the problem, why it matters, and what information or analysis might

341