Artificial intelligence is increasingly becoming part of our academic lives. From reviewing literature and improving manuscripts to exploring research questions and interpreting data, many researchers now rely on AI tools in one form or another.
Yet an important question remains: At what point does AI assistance become a substitute for independent scholarship?
I recently came across the AI policy published by Cureus, which offers an interesting and practical way of addressing this question.
Rather than imposing a blanket prohibition on AI, Cureus adopts a risk-based approach that distinguishes acceptable assistance from uses requiring greater scrutiny and practices that undermine academic integrity.
The framework identifies three categories.
Green: Permitted AI assistance
These are relatively low-risk activities, such as improving grammar, organising manuscripts, translating text, formatting documents and cleaning data.
The defining feature is that AI supports the presentation or organisation of scholarly work without replacing the researcher’s intellectual contribution.
Amber: Permitted, but requiring caution
This category covers more substantive applications, including suggesting research methods, comparing findings with existing literature, interpreting analytical outputs and providing extensive writing support.
These activities may influence scholarly reasoning. They therefore require independent verification, transparent disclosure and demonstrable human oversight.
Red: Unacceptable AI use
The framework prohibits practices that compromise scholarly integrity or replace accountable human judgment.
Examples include fabricated references or data, undisclosed AI-generated research conclusions, assigning authorship to AI, delegating peer review to AI systems and compromising manuscript confidentiality.
What does this mean for our universities?
I believe this framework deserves serious consideration by universities, academic journals and research institutions, particularly as AI-assisted research becomes increasingly common.
In legal scholarship, for example, AI may suggest possible arguments or help organise an analysis. However, verifying judicial authorities, interpreting legislation and defending legal conclusions remain the researcher’s responsibility.
The same principle applies across medicine, the sciences, humanities and social sciences.
We should also recognise that not every use of AI carries the same ethical risk. Correcting grammar is fundamentally different from allowing an AI system to generate an entire legal argument or interpret research findings without meaningful scrutiny.
Our institutional policies ought to reflect these distinctions.
A proposal for academic institutions
My suggestion is that universities and scholarly publishers consider adopting similar risk-based guidelines, appropriately adapted to their disciplines and institutional circumstances.
Such guidelines should promote transparency, protect confidential research materials, establish clear expectations for disclosure and preserve independent scholarly judgment.
Importantly, they should also educate researchers about responsible AI use rather than relying exclusively on detection and punishment.
A final thought
The academic community should move beyond asking simply whether AI was used in producing a manuscript.
We should instead ask:
- How was AI used?
- Was its contribution transparent?
- Were its outputs independently verified?
- Did the researcher retain genuine intellectual responsibility?
Ultimately, academic integrity is not protected by rejecting useful technologies. It is protected by ensuring that those technologies do not displace the reasoning, originality and accountability that give scholarship its value.
AI should support scholarship, not replace the scholar.
Source: Cureus, A Risk-Assessment Framework for Responsible AI Use in Research Publishing ↗.
This commentary represents the author’s reflections on the Cureus policy and its possible wider application. The policy itself applies to Cureus; its adoption by other institutions is proposed for consideration.
Loading reader responses…