University of Sheffield: Generative AI Principles for Research and Innovation
Source: University of Sheffield Research Ethics — Generative AI Principles Published: Current Publisher: University of Sheffield
Overview
The University of Sheffield has established five overarching principles to guide researchers in the judicious and appropriate use of Generative AI (GenAI) in their research and innovation activities. These principles sit alongside the University’s Good Research and Innovation Practices (GRIP) policy and apply to all researchers, regardless of discipline.
The focus is on integrating GenAI opportunities whilst maintaining research rigour and integrity. The University acknowledges both the transformative potential of GenAI (from productivity aids to results generation) and the emerging concerns around sustainability and ethics, emphasizing that researchers must be able to justify and explain their use of these tools.
The Five Principles
1. Appropriateness
Key question: Is this an appropriate use of GenAI?
Researchers must be able to explain and justify every use of GenAI in their research. GenAI should complement human expertise, critical thinking, and judgment rather than replace them.
Considerations:
- Does the GenAI tool(s) meaningfully contribute to the research process?
- Have you considered the sustainability implications of using GenAI?
- Are you using it only when offering significant value and contributing to meaningful, sustainable outcomes?
Implication: GenAI is not a default solution; it is a targeted intervention requiring explicit justification for each use case.
2. Attribution
Key requirement: Transparent acknowledgement of GenAI use at all points in the research and writing process.
Specific guidance:
- Include a statement such as: “This manuscript was prepared with the assistance of Generative AI [cite the tool]”
- For verbatim reproduction of GenAI output, explicitly indicate the section(s): “The Abstract and Introduction were produced by Generative AI [cite the tool]”
- Record the tool name, version, date of use, and whether the output was validated
- Keep records of the prompt texts used
- Do not use GenAI to create synthetic data for presentation in a research study without attribution
Implication: Transparency prevents misrepresentation of authorship and maintains trust in the research record.
3. Accuracy
Key requirement: Critical verification of all GenAI output before use in research.
Known risks:
- GenAI is prone to “hallucinating” and “confabulating” text, imagery, and other outputs
- GenAI systems make programming errors
- References generated by GenAI may not be reliable
- Syntheses of prior research may be biased or inaccurate
- Attributions of fact should not be trusted without independent verification
Researcher responsibility: Pay particular attention to any references, syntheses of prior research, and factual attributions generated by GenAI, verifying accuracy against authoritative sources.
4. Adherence to Data Protection Principles
Key risk: GenAI comes with risk of leakage or loss of control over sensitive and proprietary information to third parties.
Critical safeguards:
- Avoid all exposure of sensitive data (proprietary research data, personal data, etc.) to public GenAI tools
- Some GenAI tools use user data for further training of the model; check the Terms and Conditions
- Select tools with appropriate data safeguarding:
- Tools that do not use data for training at all, OR
- Tools that only train on a private version of the tool for that user
- Do not use a third party’s data (e.g., industrial partner data) in GenAI without explicit permission
- Consider export control sensitivities and other regulatory constraints
GDPR and data protection compliance: All use of GenAI must comply with UK data protection law and the University’s data protection policies. Researchers should refer to the Information Commissioner’s Office (ICO) Guidance on AI and Data Protection.
Implication: Data sensitivity assessment must precede any GenAI tool selection.
5. Accountability
Key principle: Researchers remain accountable for their research when using GenAI, regardless of the tool’s role.
Requirements:
- Ensure that GenAI use complies with ethical standards and the tenets of research integrity (GRIP policy)
- Comply with the policies of relevant funders and publishers regarding AI use
- Stay up to date with developments in the GenAI tools you are using — the landscape is rapidly evolving
Implication: Responsibility cannot be delegated to the tool; researchers must maintain continuous oversight and awareness.
Implications for this project
The five principles directly inform how tools in the catalogue should be assessed and presented to researchers:
| Principle | Catalogue relevance |
|---|---|
| Appropriateness | Tools flagged as narrow-use or discipline-specific rather than general-purpose; sustainability notes in tool metadata |
| Attribution | licence field should clarify whether academic attribution is required; note if tool’s Terms & Conditions impose restrictions |
| Accuracy | known-limitations, hallucination-risk, and output-quality-notes fields critical for research use; high weighting in assess/trial decisions |
| Adherence to data protection | data-training-policy and gdpr-compliant fields are primary procurement criteria; data-residency essential for sensitive data research |
| Accountability | funder-acceptable, publisher-acceptable flags help researchers evaluate compliance; institutional endorsement carries weight |
Related notes
- cmi-strategy-2026 — the University’s wider Centre for Machine Intelligence strategy these research principles sit within
- ukcori-uk-institutional-ai-guidance-2026 — sector-wide comparator; finds Sheffield’s Attribution/Accuracy/Accountability framing consistent with the emerging UK institutional consensus