A rapid rise in “agentic” AI—autonomous tools built on large language models that can search, draft, iterate and even submit grant applications—threatens to overwhelm traditional funding systems, argue two U.K. research leaders in Nature. Application volumes climbed sharply at a dozen major funders from 2022 to 2025, with increases ranging from 14% to 142%. At the same time, average proposal quality appears to be improving: in 2025, only 5% of EU Marie Skłodowska-Curie postdoc applications fell below quality thresholds, down from 20% in 2018. Surveys suggest AI adoption is widespread, with 58% of researchers using AI tools in 2025 and 41% deploying them for grant drafts.
Regulators are pushing back. The U.S. National Institutes of Health said in 2025 it would deem substantially AI-generated proposals ineligible, and UK Research and Innovation bars reviewers from using generative AI on submissions. The authors say such bans are hard to enforce and risk disadvantaging non-native English speakers. As both applications and reviews become AI-mediated, funding may reward formulaic conformity to past successes.
They propose shifting away from text-centric evaluations toward funding people and teams: track-record verification over time, portfolio assessments, and interviews. The UK Medical Research Council recently reinstated interviews for all shortlisted grants to blunt AI-augmented proposals.
Related articles:
San Francisco Declaration on Research Assessment (DORA)
Marie Skłodowska-Curie Actions (European Commission)
