Establishing the Validity and Ethical Legitimacy of AI-Generated Text Detection Software in Postgraduate Education: A Systematic Review of False Positives, Institutional Accountability, and Academic Justice

Abstract Introduction: AI-generated text detection software is increasingly discussed as a means of protecting academic integrity, yet the supplied evidence raises substantial concerns about validity, linguistic bias, opacity and procedural fairness. These concerns are especially important in postgraduate education, where written work may be produced by multilingual students and may carry significant progression, assessment and disciplinary consequences. Methods The researcher-selected databases and information sources were Google Scholar, ERIC, OpenAlex, Crossref. The eligibility protocol restricted publication years to 2022–2026. COSTAQDA recorded source-level retrieval through OpenAlex (38 retrieved; 38 available), Crossref (75 retrieved; 3550 available). Where researcher-selected sources did not expose a suitable direct bulk-search connector, COSTAQDA used its OpenAlex/Crossref discovery bridge while preserving the researcher-selected source names in the PRISMA record. The identification run recorded 113 records before cross-source deduplication and 111 unique records after deduplication. Results Detection performance was conditional on tool, dataset, model, task, language background, text length and evaluation protocol. False-positive risks were particularly evident for non-native and second-language English writers and professionally edited texts. A detector score did not establish authorship, intent or misconduct, especially where outputs were binary, scalar or poorly explained. Ethical legitimacy depended on human review, transparent rules, local validation, staff training, appeal processes and evidence of learning beyond the submitted text. Discussion/Conclusion: The evidence does not support treating AI-detector outputs as conclusive evidence in high-stakes postgraduate decisions. Their responsible use, where retained, requires cautious screening rather than adjudication, subgroup validation, transparent governance and meaningful human due process.

Authors

Institutions

Publication Details

Journal
Research Square
Published
2026-11-17
DOI
https://doi.org/10.21203/rs.3.rs-11731937/v1
Primary Topic
Academic integrity and plagiarism
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Establishing the Validity and Ethical Legitimacy of AI-Generated Text Detection Software in Postgraduate Education: A Systematic Review of False Positives, Institutional Accountability, and Academic Justice

King Costa, Nisi Thusi, Letlhogonolo Nolo Mfolo, Thandoluhle Tia Thusi
4 links
Research Square
Academic integrity and plagiarism
preprint

Establishing the Validity and Ethical Legitimacy of AI-Generated Text Detection Software in Postgraduate Education: A Systematic Review of False Positives, Institutional Accountability, and Academic Justice

King Costa, Nisi Thusi, Letlhogonolo Nolo Mfolo, Thandoluhle Tia Thusi
preprint en

Abstract

Abstract Introduction: AI-generated text detection software is increasingly discussed as a means of protecting academic integrity, yet the supplied evidence raises substantial concerns about validity, linguistic bias, opacity and procedural fairness. These concerns are especially important in postgraduate education, where written work may be produced by multilingual students and may carry significant progression, assessment and disciplinary consequences. Methods The researcher-selected databases and information sources were Google Scholar, ERIC, OpenAlex, Crossref. The eligibility protocol restricted publication years to 2022–2026. COSTAQDA recorded source-level retrieval through OpenAlex (38 retrieved; 38 available), Crossref (75 retrieved; 3550 available). Where researcher-selected sources did not expose a suitable direct bulk-search connector, COSTAQDA used its OpenAlex/Crossref discovery bridge while preserving the researcher-selected source names in the PRISMA record. The identification run recorded 113 records before cross-source deduplication and 111 unique records after deduplication. Results Detection performance was conditional on tool, dataset, model, task, language background, text length and evaluation protocol. False-positive risks were particularly evident for non-native and second-language English writers and professionally edited texts. A detector score did not establish authorship, intent or misconduct, especially where outputs were binary, scalar or poorly explained. Ethical legitimacy depended on human review, transparent rules, local validation, staff training, appeal processes and evidence of learning beyond the submitted text. Discussion/Conclusion: The evidence does not support treating AI-detector outputs as conclusive evidence in high-stakes postgraduate decisions. Their responsible use, where retained, requires cautious screening rather than adjudication, subgroup validation, transparent governance and meaningful human due process.

Research Square
University of Johannesburg (ZA), Adventist University of the Philippines (PH), Global Centre for Academic Research (ZA)
Quality education
Academic integrity and plagiarism
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon