EU AI funding called "out of date" - researchers push for flexibility
05 Nov 2025
EU research funding is struggling to keep pace with AI. Research commissioner Ekaterina Zaharieva said she hears regular complaints that overly prescriptive calls lock projects into plans that age before the grant is signed. "Especially now with AI, the science develops so fast that it's outdated at the moment they sign the agreement…so we should give them the possibility to do better," she told MEPs.
The system is complex, not broken
Serge Belongie, professor of computer science at the University of Copenhagen and director of the Danish Pioneer Centre for AI, said the picture "is more complex than it may first appear". Multiple EU instruments support AI with different processes and timelines. Inefficiencies happen, but he points to fragmentation across programmes rather than one systemic failure.
Progress, he argues, requires effort on both sides. Researchers should organize across institutions, form thematic teams on concrete AI challenges, and keep lines open with funders. He highlighted ELLIS as a working model that aggregates proposals, aligns institutions, and helps funders spot high-impact ideas faster.
Even open calls feel the pace
The European Research Council lets investigators pitch ideas without predefined topics, which reduces the risk of stale questions. Still, AI moves fast enough to stretch even that model. ERC grantees can face a year from proposal to acceptance and five years of execution, during which the field can change direction.
Yisroel Mirsky, a cybersecurity researcher at Ben-Gurion University, planned flexibility into his ERC Starting Grant on AI safety. He noted the pressure to move quickly and that earlier ideas can lose practical value as the field advances. He remains optimistic, saying ERC provides enough room for modest adaptations as work progresses.
Commission response: fewer rules, faster decisions
A Commission spokesperson said Zaharieva is working to simplify access to funding with a shorter, less prescriptive, and more transparent work programme. The aim is to streamline procedures and cut administrative burden by reducing topic prescriptiveness and simplifying applications. Initial steps are underway, with further changes planned.
What this means for your next proposal
- Architect for change: write adaptive work plans with pivot criteria, optional tasks, and decision gates tied to measurable outcomes.
- Focus on capabilities, not components: define milestones by performance targets and risk reduction, not specific models, datasets, or toolchains.
- Form thematic teams early: coordinate across institutions and disciplines; pre-negotiate IP, data-sharing, and compute access to avoid delays.
- Budget for iteration: include funds for model retraining, data refreshes, and evaluation updates as baselines shift.
- Prefer open or rolling instruments when they fit: investigator-led calls can give you room to adapt, but validate scope and timing against your risks.
- Maintain a living risk register: propose lightweight amendment paths with clear triggers so scope updates stay within objectives.
What funders could do next
- Write outcome-based topics with safety and evaluation constraints, not method prescriptions.
- Add rapid amendment tracks for scope updates within the same objectives and budget envelope.
- Back coordination layers (e.g., thematic networks) that aggregate proposals and reduce duplication.
- Publish decision timelines and acceptance windows; enable partial starts for low-risk work ahead of the Grant Agreement.
Keep your skills current
If your lab needs to stay current between call publication and project start, build a standing learning plan. A curated list of up-to-date courses can help fill gaps fast: Latest AI courses.
Your membership also unlocks:
