Academic integrity is an accountability system. It asks where an idea came from, who contributed to it, and who remains answerable for its accuracy and integrity. Plagiarism, AI-assisted writing and citation practice meet at that shared record of origin, contribution and responsibility.
What Plagiarism Is and What Counts as Unattributed Reuse
Plagiarism is the unattributed reuse of another person’s intellectual work in a way that presents it as one’s own. It can involve wording, ideas, arguments, methods, data or results. Reuse is not automatically plagiarism: scholarly work routinely builds on existing knowledge. The problem arises when the original source is not identified and the reader is consequently given a false account of who produced the material.
Changing wording does not necessarily answer the attribution question. If another person supplied the substance of an argument, method or result, merely paraphrasing that contribution does not make it the submitter’s own. Quotation marks alone may likewise be insufficient if the source remains unidentified.
This is fundamentally an attribution and accountability problem. Attribution connects a claim or contribution to its origin. Accountability establishes who is responsible for checking that material, responding to questions about it and correcting problems. Unattributed reuse weakens both.
What the ICMJE criteria reveal
The ICMJE Recommendations, “Defining the Role of Authors and Contributors” describe authorship through four criteria. A qualifying author must provide:
- Substantial contributions to the conception or design of the work, or to the acquisition, analysis or interpretation of data.
- Drafting of the work, or critical review for important intellectual content.
- Final approval of the version to be published.
- Agreement to be accountable for all aspects of the work, including questions about the accuracy or integrity of its parts.
These are not universal rules for deciding when a sentence needs a citation. They do, however, make the underlying logic clear: intellectual contribution and accountability must remain connected.
The ICMJE also notes that authorship alone does not communicate what qualified each named participant to be an author. Some journals therefore request and publish contribution information, at least for original research. A contribution statement records the division of intellectual work: who designed the study, analysed or interpreted data, drafted or critically reviewed the manuscript, approved the published version and accepted responsibility for it.
Where AI-Assisted Writing Sits Under Current Policy
For research publishing, the relevant reference point is Springer Nature’s Artificial Intelligence risk-assessment framework. It treats AI as a supporting technology: scholarly judgement, accountability and responsibility remain human. Its statement that “human accountability is non-transferable” means that responsibility for scholarly content, evaluation and editorial decisions cannot be assigned to an AI system.
The distinction between the policy’s green, amber and red categories turns on the degree to which a tool may influence expression, interpretation or scholarly judgement.
Green: assistive use
Green use supports expression, organisation or efficiency without influencing scientific, scholarly or evaluative judgement. The Springer Nature framework permits uses such as:
- Polishing or refining language.
- Suggesting the structure or formatting of manuscript sections.
- Translation.
- Structuring or clarifying reviewer comments.
- Comparing methodological options.
- Stress-testing research questions.
- Data cleaning and deduplication.
Permission does not remove the author’s responsibility. The framework says disclosure enhances trust and transparency by demonstrating human accountability. Even assistive output still enters work for which the author remains fully accountable.
Amber: caution under human control
Amber use may influence interpretation, framing, emphasis or evaluative judgement, but it remains under human control. Examples include:
- Suggesting analytical, experimental or methodological approaches.
- Drafting explanatory summaries.
- Comparing results with existing literature.
- Extensive copy-editing or writing support.
- Identifying patterns in exploratory data analysis.
- Recommending statistical tests or modelling approaches.
The framework permits these uses with human oversight, verification and disclosure. The relevant question is not simply whether AI was used for writing. It is whether the tool helped shape the argument, interpretation or evaluative judgement, and whether a qualified person examined and accepted that influence.
Red: not permitted
The red category includes:
- Generating hypotheses, analyses or conclusions and presenting them as human-derived.
- Fabricating data, citations or results.
- Generating core research reasoning without disclosure.
- Assigning authorship or accountability to AI systems or tools.
- Delegating peer review to a large language model.
Disclosure cannot automatically convert a red-category activity into permitted use. If an AI system has produced core reasoning, the issue is not cured by adding a disclosure statement after the fact. The activity must be assessed against the policy rather than concealed.
The continuing author duty
Springer Nature places two duties directly on authors: remaining fully accountable for originality, accuracy and integrity, and disclosing AI use. Disclosure records the tool’s role; it does not transfer judgement to the tool. The human author must still verify sources, examine methods, assess outputs and take responsibility for the submitted work.
How Citation Practice Connects Plagiarism, AI Use and Accountability
Citation practice and contribution reporting apply the same attribution logic to different questions. A citation identifies the source of intellectual material. A contribution statement identifies who performed which scholarly roles. An AI-use disclosure identifies where an automated tool entered the work. None replaces another.
Good citation practice begins before submission: the author checks that a cited work exists and actually supports the claim attributed to it. An AI-generated reference is not verified merely because it is formatted like a reference. Fabricating citations is expressly in Springer Nature’s red category, alongside fabricating data or results.
AI disclosure likewise does not repair defective citation practice. Disclosing that a tool helped rewrite or summarise material does not identify an omitted source. Conversely, a complete bibliography does not make undisclosed or prohibited AI use acceptable. The author must maintain the source trail, disclose relevant tool use and remain accountable for both.
At the authorship level, contribution statements perform a related function. Citation practice attributes ideas and evidence to sources; contribution statements attribute intellectual work and responsibility to people. Together with transparent AI-use reporting, they make the research process more checkable.
Why Rules Differ and Must Be Checked Locally
There is no defensible basis for treating one institution’s policy, one course brief or one publisher’s framework as a universal rulebook. Permitted uses, disclosure requirements, assessment designs and citation expectations can differ by context.
The TEQSA Academic Integrity Toolkit is explicit about its own status. Its resources and case studies represent approaches and ideas that institutions have found useful. They are not intended as guidance from TEQSA; they are shared to encourage institutions to consider different approaches.
TEQSA’s resource on risks to academic integrity from AI examines the detection of plagiarism in gen AI-derived text, assessment design in the age of gen AI, and ethical approaches to integrating gen AI into the curriculum. Detection is therefore only one part of a wider institutional problem. A detection label does not by itself establish who supplied particular words, who made the scholarly contribution or who was accountable.
Before using AI or submitting work, readers should check the rules that actually apply to their institution, course, assessment or publication venue. The relevant questions are concrete:
- Is the proposed use treated as assistive, interpretative or central to the reasoning?
- Must AI use be disclosed, and in what form?
- What source, citation and contribution records are required?
- What verification and human oversight are expected?
- Does a local rule differ from the framework used in research publishing?
If the written policy is unclear, clarification should be sought before submission rather than inferred from another person’s use or from a general description of “AI-assisted writing”. A defensible workflow is straightforward: keep an accurate source trail, disclose relevant AI involvement, verify outputs, record human contributions and check the applicable local rules.