The Netherlands Organisation for Scientific Research (NWO) has updated its policy on generative AI (GAI). This policy clarifies for applicants, reviewers, and staff how they should handle GAI. While applicants are allowed to use GAI, its use remains completely prohibited for reviewers and involved NWO employees within the grant process.

At the beginning of last year, NWO shared guidelines for the use of GAI. To quickly translate developments around AI and experiences with the policy into guidelines, the policy is regularly updated.

Use of GAI in the review process still prohibited

The use of GAI remains fully excluded during the evaluation of an application. Both reviewers (including referees, committee and jury members) and NWO employees must comply with confidentiality obligations for all documents related to an application during the grant process. NWO cannot guarantee this confidentiality when GAI is used by reviewers and NWO staff involved in the review process. Additionally, NWO has a duty of verification. This means NWO must be able to check whether a reviewers advice is based on a careful assessment. The use of GAI complicates compliance with this duty, as it is often not possible to trace how GAI was used. 

Key updates

The aim of the update is to simplify the guidelines where possible and add guidelines for new developments or gaps where necessary. Below is an overview of the main updates.

General policy framework:

  • The unreliability of AI control software is included in the policy framework. Currently, such applications are not reliable enough to establish the use of GAI and are therefore not used by NWO. Results from such applications are not accepted as proof of GAI use.
  • The principle regarding sustainability has been simplified to clarify how NWO deals with this and what advice is given to applicants, reviewers, and NWO staff. For example, NWO staff are asked to check whether the use of GAI is necessary or if a more sustainable solution is available.

Applicant guideline:

  • The guideline for applicants has been expanded with a ban on including hidden prompts in their application. A hidden prompt is an instruction to a GAI application that is not (or poorly) visible to the reader of a document, intended to influence the GAI application with a new command. Such a command may be to generate a positive assessment of the relevant preceding text. This is not allowed under any circumstances.
  • The guideline regarding transparency has been simplified; applicants are only asked to mention GAI as a source in the source citation of their application if GAI is used.

Reviewer and/or referee guideline

  • The ISAAC form for referees includes a declaration that the referee does not use GAI.
  • It is explicitly stated in this guideline that the use of GAI as a language aid or for translations is not allowed for referees and reviewers.

NWO staff guideline

  • This guideline includes how NWO staff should deal with partners who use (or want to use) GAI. In such situations, NWO staff must check whether GAI was actually used and, if so, prevent confidential information from being shared.

Use of GAI for NWO staff in other tasks

NWO staff may use GAI for tasks outside the review process, but only GAI applications that have been approved and declared safe according to the (IT) procurement process and made available. Currently, NWO staff are allowed to use Microsoft Copilot Chat and the DeepL translation application.

Read the policy document