One of my NIH grant reviews came back last year with comments clearly LLM-generated. The PMIDs cited against the proposal were hallucinated. The papers did not exist. The score still counted toward the funding decision.

A new Lancet analysis puts numbers to what I saw: fabricated references in biomedical papers are up 12-fold in two years. 1 in 277 PubMed-indexed papers this year cites work that does not exist.

The reflex is more disclaimers: AI-use statements, acknowledgments at the bottom of reviews. Three years into that experiment, the trajectory has only steepened. It is not working.

Punitive measures are easier to propose than to enforce. Reviewers cannot hand-verify every citation. Retraction does not unread a paper or unscore a grant. The damage compounds before the correction lands.

What if the path forward is to build different tools? AI that grounds every citation against PubMed before submission. AI that verifies every reference in a peer review before it reaches the program officer. AI that assists judgment rather than generates it.

I don't have a clean answer. Science will be AI-touched at every level. AI-assisted and AI-generated are not the same thing. What does the first actually look like in practice, and who should be building it? thelancet.com/journals/lance…

Technology to verify citations is easy and numerous tools already exist. They just have to be implemented. It would be trivial to have one or several AI models do initial screening for citations (both meta-data and appropriateness) and whatever else was high yield. Not to mention, we could just require doi’s for every citation and have an editorial assistant click on them. That would take 2 minutes.