AI agents pose existential threat to research funding

Increasingly advanced AI agents are profoundly changing how researchers develop grant proposals and how reviewers assess them. While these technologies offer exciting possibilities, they also pose severe structural challenges to the global research funding ecosystem.

On Friday, 26 June, RoRI’s Executive Director James Wilsdon joined Geraint Rees (Vice-Provost for Research, Innovation and Global Engagement at UCL) for an online seminar hosted by the League of European Research Universities (LERU). Building on their recent analysis in Nature, which utilised RoRI data across 12 global funders to reveal a 57% spike in applications between the launch of ChatGPT and late 2025, the session engaged an audience of approximately 400 university leaders, funders, and research policy actors to discuss the rapid rise of agentic AI.

The seminar was covered by Times Higher Education, highlighting the critical shift from standard generative AI to autonomous AI agents that can independently write and submit proposals, effectively driving the marginal cost of applications down to zero.

As AI optimises writing, the “quality floor” rises while the ceiling stays the same, making it increasingly difficult for reviewers to discriminate between truly excellent ideas and average ones.

A near-future risk where AI agents end up writing proposals that are then evaluated by AI reviewers trained on the exact same data, leaving the system to measure simulation rather than genuine human ideas.

Wilsdon and Rees argued that the architecture of the current system is no longer fit for purpose, and future evaluation must shift toward what agentic AI cannot simulate, such as a researcher’s track record of delivering great ideas.