In four recent opinion articles, Nature questions the response of the European Research Council (ERC) to a steep increase in grant applications supported by AI tools, namely Responses to the AI grant flood must prioritize fairness as part of excellence, Prestigious European science funder scraps stricter rules after researcher backlash, Could agentic AI topple grant-funding systems?, and European funder must increase capacity to meet the ambition of scientists,
What is happening is that LLMs are making the construction and design of grant applications much easier, even improving their scientific worth in many cases. Hence seeing the number of submitted proposals rise. For instance, an 142% for EU Marie Skłodowska-Curie Actions fellowships. This is getting worse with agentic AI. The funding bodies are unable to implement the various bans they set against relying on these agents. The low-tech solution of increasing the role of interviews and team assessments, while worthwhile, is not working above a certain threshold, being time-intensive (and open to many biases). A comment of 27 April 2026 suggests burning all bridges and resorting to locally designed agentic AIs to assess these proposals… As if the well-studied drawbacks of AIs and LLMs relying on black-box, biased, bibliometric, unsecured, machines would magically vanish with this new construct! The mention of post hoc human verification is also wishful thinking, considering the impact of earlier bibliometric tools such as the h-factor. (Even more recent proposals are similarly bananas.)
An editorial of 05 May 2026 claims the higher moral ground by calling for fairness and inclusivity. But with a bounded budget and an equally bounded pool of human reviewers, the protocol of evaluating an ever-growing pile of applications cannot remain the same. The ERC step of limiting how many grants researchers can apply for and how long unsuccessful candidates have to wait before they can re-apply has been rescinded under protest from some researchers. But a call of 01 May 2026 for more structural investment does not stand the test of reality. Working in an ERC panel is time-consuming, equivalent to one month of investment, and finding more willing experts is unlikely to succeed, not mentioning the rise in the ERC budget the EU Commission is not going to vote, the more (or the less!) with the rise of populist fractions in the European Parliament
Chapter 5 addresses the crucial challenge of ethical evaluation in a rather verbose and vague manner. For instance, with no instruction on how to resist adversarial attacks. Or criticising p-hacking and multiple testing while missing the elephant in the room (
