AI for Research Ethics: Authorship, Disclosure, and Data Boundaries

AI for research ethics covers the decisions a researcher must make before using AI in a study: whether AI assistance affects authorship, what must be disclosed in a paper or grant application, which participant data can enter an AI tool, and whether the institution needs to approve the approach. These are not hypothetical concerns—journals, funders, and IRBs are already requiring answers to them. fenbs helps research teams record institution-approved rules, disclosure decisions, and their rationale so every team member works from the same agreed boundaries.

The guidance is evolving quickly. Many journals updated their AI policies in the past two years. The safest approach is to resolve each of the questions below before beginning the research, not when revising a manuscript.

What changes when AI assists with research

AI assistance does not change the researcher's obligation to ensure that every claim is supported by evidence, every citation is accurate, and the methods are reported completely. What it changes is the set of additional questions that arise: Was AI used in the analysis? In the writing? In the literature search? Did participant data enter the tool?

Each of those questions has a different answer and a different obligation. Researchers who treat AI as a single undifferentiated category—"we used AI tools"—often end up with disclosure statements that satisfy nobody. Think task by task: what did the AI do in this study, and what does that mean for this journal's policy, this funder's requirements, or this IRB's scope?

Authorship and credit questions

Current guidance from major journals and research bodies holds that AI tools cannot be authors. The most widely cited position is from ICMJE (International Committee of Medical Journal Editors), which states that AI tools cannot meet the criteria for authorship—they cannot take responsibility for the work, they cannot give informed consent, and they cannot be held accountable. Check the ICMJE guidelines at https://www.icmje.org/↗ (opens in a new tab) before submitting to any medical or life sciences journal.

What does change is the human author's responsibility. If an AI tool substantially drafted text, performed an analysis, or generated figures, the corresponding author is responsible for reviewing and verifying that output. Listing AI as a contributor in the methods section is increasingly required by journals that previously only asked for disclosure. Check the specific journal's current policy—it may have changed since a co-author last looked.

Disclosure obligations

Most journals now require disclosure of AI tool use somewhere in the manuscript. The location varies: methods section, acknowledgments, or a dedicated AI use statement. Funders including NIH and NSF have their own disclosure or restriction policies; check the specific program announcement, not just the funder's general AI page.

A disclosure that will satisfy editors typically names the tool, describes the task it performed, and states what human review followed. "We used ChatGPT for some writing tasks" is likely insufficient. "We used GPT-4 to generate initial summaries of included papers; all summaries were reviewed and rewritten by the authors" is more likely to satisfy current policies.

Protecting participant data

Before entering any participant data into an AI tool, ask: is this tool covered by your institution's data security agreement? Does the IRB consent form authorize this use? If participant data is protected under HIPAA, FERPA, or a consent form that specifies how the data will be used, running it through an unapproved third-party AI tool may be a breach of that consent or of applicable law.

Most AI tools send data to external servers for processing. Enterprise or confidential-processing agreements can limit this in some tools, but those agreements must be in place before the data enters the tool, not added afterward. Work with your IRB and institutional research security office on approved tools and data categories before analyzing participant data with AI.

Getting institutional approval

Depending on the study type and your institution, you may need IRB review, data security review, or both before using AI in the research workflow. This is especially true when: the AI tool will process participant data; the AI will be used in the analysis or interpretation of results; or the tool's outputs will influence decisions that affect participants.

Document the approval and its scope. If the IRB approved AI use for literature screening but not for participant data analysis, that distinction needs to be in the record. Approval for one task does not cover all tasks, and scope creep in AI use is as much a problem in research as in organizational settings.

How to do this with fenbs

Set up the governance record before the study begins, so every team member starts from the same approved rules.

1. Create a free account and set up a project for the study. Add a short description of the research goal and AI tools under consideration.
2. Write the approved AI rules in Decisions and rules: permitted tools, data categories off-limits, required disclosure language, and who approved each decision. Name the IRB or department head who authorized it.
3. Create a task for each team member who will use AI in their part of the research. Add a checklist: approved tool only, no participant data, review completed.
4. When journal or funder requirements arrive, add a decision for the specific disclosure obligation so the team has the exact wording required.
5. Before submission, save a progress note with the disclosure statement text, the tools used, and the tasks completed. Future team members or co-authors can see the full record.

fenbs does not connect to research databases, IRB systems, or grant management platforms. It holds the governance layer: approved rules, team decisions, and the rationale behind them. For questions about keeping AI work records outside the model, the guide to what is an AI brain covers the basics.