AI-drafted grants edge NIH funding but risk dampening novelty

AI-assisted NIH grant proposals appear about 4% more likely to be funded and yield more papers, but they tend to resemble prior work, raising concerns about reduced novelty and potential homogeneity; NSF proposals show no funding advantage from AI use. Findings come from an arXiv preprint analyzing NIH/NSF submissions and AI-rewritten abstracts, and are not yet peer reviewed.