When AI turns grant evaluation into a lottery

Research funding schemes are increasingly struggling to reliably distinguish scientific merit through traditional scoring. Using the most recent evaluations of the EU Marie Skłodowska-Curie Actions postdoctoral fellowships as a case study, we show how the rapid institutional adoption of Large Language Models coincides with unprecedented score compression. With only ~5% of proposals now falling below the 70% quality threshold, down from ~20% in previous years. We argue that “excellence saturation” has reached a tipping point that exposes the structural limits of fine-grained peer review and alters reviewer decision-making dynamics where funding decisions resemble a lottery. This shift to AI-assisted grant writing effectively decouples a proposal’s form from its scientific substance, necessitating a transition from fine-grained ranking toward managing an abundance of excellence through alternative allocation mechanisms, such as funding lotteries.