Human analysis and the expert assessment of the proposal are core elements of the review process. The use of AI systems in the preparation of reviews is therefore permitted only in a supporting capacity. All substantive judgements regarding the funding recommendation rest solely with the reviewer.

AI systems may be used only where care is taken to ensure that confidential content of funding proposals is not processed without oversight, stored, or used for purposes other than the conduct of the review process. The supportive use of AI in drafting a review must not result in the disclosure of confidential proposal content. This can be ensured by using the following types of system:

1. Locally hosted systems without cloud computing
2. Systems hosted by a trustworthy institution (e.g. a university or public research institution)
3. Cloud-based (commercial) systems subject to a contractual licence agreement guaranteeing data security

If an AI system is used in preparing the review, this must be disclosed, specifying the purposes of use.

Uncritical adoption of AI-generated text is not permitted. AI-generated text must always be critically assessed, adapted and checked for factual accuracy, currency and potential bias structures.

Full responsibility for the content of the review lies with the person who prepared it. This applies even where parts of the text were generated with the assistance of AI systems. Responsible expert review requires subject expertise and ethical judgement, which by their very nature AI systems do not possess. Responsibility must therefore not be transferred to an AI system; it remains with the reviewer.

The digital turn is also impacting on the way research is conducted, including its quality assurance. AI systems offer new possibilities for information processing and text support which, when used responsibly, can help ease workloads. For this reason, the DFG has decided to permit the use of AI systems in the review process – subject to clearly defined conditions. The underlying principles here are confidentiality, transparency, quality assurance and responsibility. These are set out in detail in a binding guideline on the use of AI in the review process.

AI systems may only be used in a supporting capacity. This includes, for example:

• searching for relevant academic literature,
• linguistic or structural revision of one’s own draft texts,
• expanding one’s own bullet points or notes,
• formal checks (e.g. clarity, readability).

Human analysis and the expert assessment of the proposal are core elements of the review process. The use of AI systems in the preparation of reviews is therefore permitted only in a supporting capacity. All substantive judgements regarding the funding recommendation rest solely with the reviewer. Further details are set out in the Guideline on the Use of AI in the Review Process(interner Link). For example, it is not permissible to have substantial parts of the review generated automatically and without reflection, i.e. without independent expert assessment.

When registering in elan, you confirm that if you use AI in the review process, you have sufficient AI literacy. The DFG relies on your self-assessment in this regard. There is no absolute standard here. Key elements of AI literacy include a basic technical understanding of how AI systems function, awareness of their limitations such as hallucinations and bias, knowledge of how AI systems can be used in the review process, and, closely related to this, a clear understanding of one’s own role in the epistemic evaluation process (see also the Guideline on the Use of AI in the Review Process(interner Link)). For a useful summary of AI literacy in general, see the paper Hinweispapier KI-Kompetenzen nach Artikel 4 KI-Verordnung(externer Link) [AI literacy under Article 4 of the AI Regulation], issued by the Federal Network Agency. A wide range of free online training courses are available on the KI Campus(externer Link) website operated by Stifterverband für die Deutsche Wissenschaft e.V.

If the information transmitted to an AI system includes personal data, its use constitutes data processing within the meaning of the General Data Protection Regulation (GDPR). Since this processing is carried out on behalf of the controller, Article 28 GDPR requires the conclusion of a data processing agreement (DPA). Such agreements can and must be concluded with the respective provider as part of the licensing arrangement. Where you use an AI system licensed by your university or research institution, this is the responsibility of the institution. If you use an AI system hosted exclusively locally at your university or research institution, there is no requirement to conclude a data processing agreement (DPA), as in this case no processor is involved.