Pull to refresh
Logo
U.K. funder uses AI triage to reject half of grant proposals

U.K. funder uses AI triage to reject half of grant proposals

Rule Changes

CRANE scored 179 cybersecurity proposals with AI and cut the bottom half before human review; its parent funder says its policy forbids generative AI assessment.

Today: Backlash reaches the national funder

Overview

Updated 1 hour ago

Claire Hardaker's proposal on the risks of trusting AI in voice analysis was cut by an AI. The Lancaster University linguist was one of roughly 90 researchers rejected by CRANE, a U.K. cybersecurity funder, before any person read their applications.

CRANE scored all 179 proposals with AI models and discarded the bottom half. Funders worldwide are testing AI to handle record application volumes; this is what happens when the machine makes the first cut without explaining itself.

Why it matters

If AI triage spreads without transparency rules, a researcher's funding can hinge on an opaque algorithm that offers no appeal and no explanation.

Questions about this story

Free account needed to ask — your question is kept and asked for you right after sign-up. Answers are public.

No questions yet — be the first to ask.

Key Indicators

179
Proposals evaluated by AI triage
Total applications to CRANE's phase one pilot call for cybersecurity projects.
50%
Proposals rejected before human review
The bottom half of the AI-ranked list was discarded before any human reviewer saw it.
2
AI assessment methods used
CRANE had separate AI models rate proposals, then had models 'debate' merits before averaging the scores.
8
UKRI assessment modernization measures
Rolling out from October 2026, including a retrospective trial of AI-assisted assessment.

Voices

Curated perspectives — historical figures and your fellow readers.

Ever wondered what historical figures would say about today's headlines?

Sign up to generate historical perspectives on this story.

People Involved

Organizations Involved

Timeline

July 2026 October 2026

4 events Latest: Today
Tap a bar to jump to that date
  1. UKRI unveils assessment overhaul

    Upcoming Policy

    UKRI introduces eight measures from October 2026, including a retrospective trial of AI-assisted assessment comparing AI judgments against past human decisions.

  2. Backlash reaches the national funder

    Today Statement

    Science reports the story. UKRI says its policy forbids generative AI assessment and it is following up with CRANE to ensure compliance with its assessment conditions.

  3. AI triage rejects half of applicants

    Decision

    Rejection emails tell roughly 90 researchers their proposals failed an AI-based triage, without revealing which AIs judged them or why. (Date approximate: emails arrived the week before the Science report.)

  4. CRANE opens cybersecurity pilot call

    Funding Call

    CRANE issues its phase one pilot call; 179 proposals arrive. Guidelines say AI tools 'may' be used in initial assessment, with final ranking by the award panel. (Date approximate: proposals were drafted over the summer.)

Scenarios

1

UKRI blocks AI desk-rejection, mandates transparency

Likely Resolves by Q1 2027

Discussed by: Science, Times Higher Education

UKRI's compliance review concludes CRANE's generative AI used to reject proposals conflicts with its policy. UKRI issues a framework requiring that AI in triage be disclosed to applicants and that any rejection pass through a human. CRANE revises its process before its next call.

2

AI-assisted triage becomes standard across U.K. funders

Possible Resolves by Q2 2027

Discussed by: Nature, Research Professional News

UKRI's retrospective trial, comparing what AI would have concluded about past applications against human decisions, shows acceptable alignment. Facing surging application volumes and warnings that grant systems could overload, UKRI expands AI-assisted triage with staff-led desk rejection across research councils, with transparency rules attached.

3

CRANE reinstates human review for rejected proposals

Possible Resolves by Jan 31, 2027

Discussed by: Rejected applicants on social media

Under pressure from applicants and the parent funder, CRANE announces a formal appeal process or reopens the 179 proposals for full human review. Some rejected proposals win funding after a person reads them.

Historical Context

3 moments from history that rhyme with this story — and how they unfolded.

2014–2018

Amazon scraps AI recruiting tool (2018)

Amazon built a machine-learning tool to rank job applicants. Trained on a decade of Amazon resumes, it penalized resumes containing words like 'women's,' effectively favoring male candidates, and the company quietly abandoned it.

Then

Amazon scrapped the tool before it was ever used in real hiring decisions.

Now

It became the standard cautionary tale about AI replicating bias from historical training data.

Why this matters now

It shows a private AI gatekeeper failing and being withdrawn; the difference is that CRANE's decision involves public money and the system's workings were hidden from applicants.

August 2020

UK A-level grading algorithm (August 2020)

When the pandemic cancelled exams, the U.K.'s exams regulator Ofqual used an algorithm to standardize teacher-assessed grades. It downgraded roughly 40% of students' predictions, with the steepest cuts falling on students from disadvantaged areas and smaller schools.

Then

Within days, after street protests and a government U-turn, the algorithm was scrapped and students received their teachers' grades.

Now

The collapse damaged public trust in algorithmic decision-making in high-stakes settings and showed that opaque scoring systems can be reversed under pressure.

Why this matters now

Like CRANE's AI triage, the A-level algorithm replaced human judgment with an opaque system that could not explain its decisions, and it was abandoned when the public saw the results.

2021–present

La Caixa's AI-assisted review (2021)

Spain's La Caixa Foundation began using AI-assisted assessment for its research funding in 2021, but its guidelines state that 'no proposal will be discarded without expert human intervention.'

Then

The approach ran for years without a comparable scandal, with AI assisting rather than deciding.

Now

It stands as a working example of AI used in peer review with a human backstop.

Why this matters now

La Caixa shows a transparency-first alternative: AI can help triage, but a person always sees the proposal before it is rejected.

Sources

(6)