Purpose and position
The Nuffield Foundation recognises that generative AI tools offer significant opportunities to support research and innovation, while also introducing risks relating to research integrity, transparency, confidentiality, and bias.
This policy sets out our expectations for the responsible and ethical use of generative AI in funding applications and assessment processes.
In light of the fast-moving development of generative AI use, this policy will be subject to periodic review.
Core principles
We expect any use of generative AI to adhere to the following overarching principles:
- Integrity and originality: submissions must meet established standards of research integrity, including avoidance of fabrication, falsification, and plagiarism.
- Rigour and reliability: submissions must be critically evaluated to identify and correct inaccuracies, including hallucinated content.
- Accountability: individuals using AI remain fully responsible for their submissions.
- Confidentiality and data protection: sensitive, personal, or confidential information – including funding applications and peer reviews – must be processed securely. It is essential that only private enterprise generative AI tools are used with confidential information to prevent unauthorised use or access.
- Fairness and equity: AI use should not introduce bias or undermine fair assessment processes.
- Legal and ethical compliance: AI use must comply with intellectual property, copyright, and data protection requirements.
These principles reflect a commitment to maintaining high standards in research while enabling responsible innovation.
Applicants’ use of generative AI
In all instances, we consider applicants to be fully accountable for the quality and integrity of their submissions.
If applicants make use of generative AI they must adhere to the core principles outlined above.
However, we would advise applicants to use generative AI tools judiciously since an over reliance on generative AI can result in generic, formulaic or unconvincing applications.
We would emphasise that applicants should ensure that their use of generative AI tools does not breach standard expectations of research integrity, for example by:
- Fabricating, falsifying, manipulating, or misrepresenting data or results
- Generating fictitious citations or references
- Plagiarising or reproducing others’ work without attribution
- Breaching confidentiality, data protection, or intellectual property rights
Any suspected misuse will be investigated.
Reviewers’ use of generative AI
In all instances, we consider reviewers to be fully responsible for the originality and validity of their assessments.
We select peer reviewers for their expertise and experience. We value their unique perspectives, and the spirit of dialogue between reviewers, applicants and the Nuffield Foundation. This expertise and contribution cannot be replaced and should not be undermined by the use of generative AI.
If reviewers make use of generative AI for a specific purpose while drafting their review, we expect them to adhere to the core principles outlined above.
We would emphasise that the maintenance of confidentiality is essential for safeguarding the exchange of scientific opinions and assessments. So, only private enterprise generative AI tools are permissible1, as these will usually offer greater data security than publicly accessible AI tools. Reviewers must not upload or otherwise share content from funding applications into publicly accessible AI services that may allow unauthorised use or access, for example by using the data to train underlying models.
Nuffield Foundation’s use of generative AI
The Nuffield Foundation does not use generative AI to make funding decisions. We may sometimes use generative AI to support the assessment process. Whenever this is the case, we adhere to the core principles outlined above.
[1] For example, private enterprise generative AI tools include: Microsoft 365 Copilot deployed within an organisation’s Microsoft tenant; ChatGPT Enterprise; Google Gemini for Workspace (enterprise deployment) or an organisation’s own internally hosted large language model. Public consumer versions of ChatGPT or consumer AI apps where users operate outside their organisation’s managed environment are not permissible.
