Daily Nous Advises Referees on Suspected AI Writing
Daily Nous published guidance on July 24 for journal referees who suspect a submission was substantially or improperly written with an LLM. The article recommends pausing review, documenting a small number of concerning passages, avoiding AI detectors because of confidentiality and accuracy concerns, and escalating the matter to an editor. It also notes that disagreements between reviewers and editors can leave reviewers uncertain how to proceed.
Daily Nous published guidance on July 24 for academic referees who suspect a manuscript was substantially or improperly written using an LLM. The article advises reviewers to pause their assessment, preserve a few examples of passages that prompted concern, and contact the journal editor or managing editor rather than independently resolving the allegation.
The recommended sequence begins with uncertainty: reviewers should consider that their assessment may be wrong and that any AI use could be permitted under a journal's rules and disclosed in the submission process. Daily Nous argues that a suspicion of improper AI use can bias a referee's evaluation of the manuscript, making a pause appropriate while the editor investigates.
The article advises against submitting a confidential manuscript to an AI detector. It cites two concerns: doing so may conflict with referees' confidentiality obligations, and detector accuracy remains questionable. Instead, the suggested evidence is a short set of passages accompanied by an explanation of why each appears AI-generated.
Editorial handling remains the decision point
According to Daily Nous, an editor could investigate the concern, relieve the reviewer of the assignment, or determine that the suspicion is unfounded and ask the reviewer to complete the report. The article includes an account from an associate professor of philosophy who reported disagreeing with an editor over whether a submission was largely AI-written, then faced the task of providing author-facing comments alongside a rejection recommendation.
The issue is already producing formal submission rules in parts of the ML research ecosystem. In a June 2 post, the NeurIPS 2026 Position Paper Track chairs wrote that final papers must be substantially human-written, permitting AI only for copy-editing or similarly peripheral changes to the main text. The chairs reported that, after analyses conducted with detector vendor Pangram, 178 submissions, or 18.4% of submissions, would be desk rejected and 123 submissions, or 12.7%, would be asked to provide evidence of substantial human engagement or risk rejection.
The NeurIPS policy applies to a specific conference track, not to journal peer review generally. Still, the contrast illustrates a broader governance problem: peer-review systems need procedures that protect manuscript confidentiality, avoid overreliance on probabilistic detection tools, and assign misconduct determinations to editors rather than individual reviewers.
Key Points
- 1Daily Nous recommends that referees pause review and escalate documented concerns to editors rather than independently adjudicating suspected AI authorship.
- 2The guidance cautions against AI detectors because confidential manuscripts may be exposed and detector outputs can be inaccurate.
- 3Formal AI-authorship policies, such as NeurIPS' position-paper rules, show how review systems are creating distinct evidence and escalation processes.
Scoring Rationale
The guidance addresses a growing operational problem for academic reviewers and editors as LLM-assisted writing becomes more common. It is relevant to ML researchers who submit, review, or administer scholarly work, but it does not introduce a new model, tool, or broadly binding policy.
Sources
Primary source and supporting public references used for this report.
Practice with real Ad Tech data
90 SQL & Python problems · 15 industry datasets
250 free problems · No credit card
See all Ad Tech problems
