The rise of extra refined synthetic intelligence brokers poses an “existential risk” to the best way analysis funding is awarded, consultants have warned, saying autonomous AI programs might additional flood grant competitions with purposes, making it tougher to establish the perfect concepts.
Talking throughout a webinar organized by the League of European Analysis Universities, Geraint Rees, vice provost for analysis, innovation and world engagement at UCL, mentioned a brand new wave of AI instruments represents a elementary shift from immediately’s extensively used generative AI.
Not like giant language fashions akin to ChatGPT and Claude, AI brokers can autonomously collect data, make choices and produce work with minimal human oversight.
“Within the context of grants, generative AI could assist you to polish or write a greater software,” Rees mentioned. “However agentic AI will go off and write the applying and submit it for you.”
Such programs could be skilled on a researcher’s revealed work, funding standards and beforehand profitable grant purposes to generate, evaluation and enhance proposals.
“The marginal value of manufacturing an software falls to zero,” Rees mentioned. “That’s essential as a result of it adjustments the character of the issue. It’s not about sprucing your grant. It’s a couple of system that’s designed basically to evaluate human judgment and human concepts being fed outputs that simulate human judgment.”
Rees and James Wilsdon, a professor of analysis coverage at UCL and govt director of the Analysis on Analysis Institute, mentioned AI brokers are creating three main challenges for analysis funders—a large improve in purposes, “high quality compression” and “convergence.”
The pair’s latest analysis analyzed knowledge from 12 funders throughout seven analysis programs and located grant software volumes had elevated by 57 p.c between the launch of ChatGPT in 2022 and the tip of 2025. Early knowledge for 2026 suggests the speed of progress is accelerating much more, Wilsdon mentioned.
The second problem is what the researchers described as “high quality compression.”
“When everybody’s proposal has optimization and nice writing, the standard ground rises, however the ceiling stays the identical,” Rees mentioned. “It turns into tougher to discriminate between actually glorious concepts and ones which can be simply OK.”
The longer-term concern is convergence, the place AI programs generate grant purposes whereas additionally aiding reviewers in assessing them.
“In case you’ve acquired brokers writing, and brokers reviewing, and the brokers are speaking to one another, they’re all skilled on precisely the identical work,” Rees mentioned. “Over time, the system isn’t going to judge nice concepts. It’s simply going to measure how nicely brokers simulate what funders have beforehand rewarded. That’s an actual existential risk to what we’re attempting to do.”
The researchers warned that the velocity of technological change is more likely to outpace the power of universities and funders to adapt.
“In case you assume it’s unhealthy now, simply wait one other yr,” Rees mentioned. “The instruments are bettering at such a speedy charge that we run the danger they’ll outpace our capability to remodel the grant-funding system.”
Each researchers rejected strategies to ban AI in grant writing, saying it’s unattainable to police.
“Bans will not be enforceable,” Rees mentioned. “Makes an attempt to detect AI produce an enormous variety of false positives. It’s not a sensible method.”
The pair argued that funders and universities ought to work collectively to redesign analysis evaluation.
“At present funders are bringing short-term fixes that usually offload a few of the work to universities,” Rees mentioned. “However it’s about pondering deeply in regards to the structure of the present system, which is now not match for function due to LLMs and agentic AI.
“We want to consider evaluating based mostly on what agentic AI can’t simulate: analysis observe file as acceptable to a researcher’s profession stage … observe file of delivering nice concepts and pondering deeply about what a fantastic concept is within the first place. In order that’s going to imply a distinct emphasis in how we consider folks and grants.”
