The European Research Council (ERC) Scientific Council has issued new guidelines regarding the use of generative artificial intelligence in the grant evaluation process, emphasizing that the human element remains non-negotiable. While the guidelines allow for minor administrative assistance, they establish a firm “no-delegation” policy that prohibits AI from infringing on the professional judgment of reviewers.
Reviewers are barred from using AI to double-check their own work or identify missing points in their evaluations. Because determining what is scientifically relevant or important is considered a specialized act of judgment, delegating this task to a tool is seen as a breach of trust. Reviewers may still use AI for basic language polishing or general information gathering, provided no sensitive data is uploaded, but the intellectual heavy lifting must remain strictly human.
At the heart of these regulations is the principle of total reviewer responsibility. Even in cases where a reviewer might operate an AI model locally on their own hardware to ensure that no data can be leaked to third parties, the Council explicitly forbids using AI to summarize proposals. The ERC maintains that fully grasping a proposal is a core duty of the evaluator and relying on a machine-generated summary risks overlooking the nuances of the applicant’s original work.
