AI Use in Peer Review Persists Despite Bans, Revealing Enforcement Gaps

The New Scientist report describes an experiment where conference organizers directed reviewers to avoid AI tools during paper assessments. Despite clear prohibitions, many participants proceeded with models like GPT variants for summarization or scoring. This occurred in a setting where peer review integrity directly affects publication decisions and career outcomes for authors.

Data from the test showed that self-reported compliance was low while indirect signals such as stylistic consistency and speed suggested widespread undisclosed assistance. Related studies, including a 2023 arXiv preprint on LLM detection in reviews and a Nature survey on researcher attitudes toward generative tools, indicate similar patterns across disciplines where time pressure overrides stated rules.

Ethical implications center on eroded trust in evaluation systems and uneven advantages for reviewers with better AI access. Human behavior here follows established patterns seen in prior technology adoptions, where convenience trumps compliance absent strong detection or penalties. Original coverage underplays how this dynamic could accelerate if conferences lack verifiable audit methods.

Future enforcement will likely require hybrid detection pipelines combined with revised incentive structures rather than simple policy statements. Without such measures, undisclosed AI integration risks becoming normalized in scientific gatekeeping.