Cheating or being cheated? Academic integrity and a changing cheating habitus in international AI governance discourse

Synthesis: Hashem (2026) asks what international AI policy actually says about academic integrity in higher education, and finds that it says a great deal in a vocabulary that never settles into rules. Reading four UNESCO and OECD publications through Fairclough's critical discourse analysis, with Bourdieu's theory of practice and Laclau's floating signifiers as interpretive tools, the paper shows eight recurring ethical terms — responsibility, transparency, trust, human agency, originality, authenticity, responsible use and augmentation — carrying normative weight while shifting referent from document to document. The consequence is that no actor is left holding the obligation: responsibility is invoked by providers, institutions, teachers and learners at once. The paper names this a changing cheating habitus, proposes that it is stratified by the academic capital students bring, and argues in its sharpest move that cheating and being cheated are not alternatives but concurrent conditions. Nothing here is measured; the claims are propositions drawn from policy text.

Key Findings

1. The corpus is two organizations and four documents, read as discourse rather than as policy. UNESCO's Guidance for Generative AI in Education and Research (2023), the OECD's Digital Education Outlook 2023, UNESCO's AI and the Future of Education (2025) and the OECD's Digital Education Outlook 2026. The author is explicit that genres differ — institutional guidance, policy outlooks, and an anthology of commissioned contributions whose authors do not speak for UNESCO — and attributes anthology evidence to its chapter authors rather than treating the volume as an institutional position.
2. Responsibility is distributed until it belongs to no one. UNESCO (2023) holds providers "responsible for ensuring adherence to core values" while requiring educators to assure "responsible, pedagogically appropriate interaction" with AI tools; the OECD (2026) narrows the same term by making teachers "remain responsible for the quality of the output by evaluating, modifying or endorsing the AI-generated output." Each formulation is internally coherent; together they leave the boundary between augmentation and substitution to local interpretation, in passive and modal constructions that express expectation without an implementation mechanism.
3. Trust migrates between three incompatible referents. UNESCO (2023) states that AI outputs "cannot be trusted to be accurate" while calling for systems to become "trustable" through human-centered governance; the OECD (2026) relocates trust into a human capacity to "meta-cognitively assess when to trust AI input and when to ignore it." Trust is therefore simultaneously a property of outputs, an objective of governance and a metacognitive skill, with no shared account of what is being assessed.
4. The governance gap is documented, not transitional. The OECD (2023) reports that as of early 2024 none of the eighteen surveyed countries had issued specific regulation on generative AI in education, and that nine had delegated primary governance responsibility to the school level. The later OECD (2026) finds twenty-three countries and jurisdictions converging on the same ethical vocabulary while leaving its operational meaning to contextual interpretation. The paper reads this through Ahmed's non-performativity: institutional discourse expresses ethical responsiveness without establishing any mechanism through which the commitment could be enacted or judged.
5. Authenticity is being redefined in the corpus but not governed by it. The OECD (2026) registers a shift from the originality of final products toward process-oriented demonstrations that reveal "the authenticity and depth of their learning," and separately warns that AI-enhanced performance can produce a "mirage of false mastery" in which high-quality output conceals limited learning. The paper's point is that once the quality of a product stops being evidence of the learning behind it, the difficulty moves from enforcement to the basis on which participation is judged at all.
6. Cheating and being cheated are concurrent, not sequential. The title's question is answered structurally: a student sanctioned for AI-assisted work may be cheating in the conventional sense and simultaneously be cheated by frameworks that normalize the same tools while retaining integrity standards built for autonomous authorship. The paper adds a hypothesis about enforcement — if stronger disciplinary knowledge and prompting literacy let students revise AI output into an accepted academic register, then the visibility of AI assistance varies with academic capital, and detection-led governance functions as a filter for that capital rather than as a neutral instrument.
7. The habitus is proposed as a relational explanation, not a typology of students. Students whose dispositions align with recognized academic expectations can treat AI as an extension of judgment they already have; students whose resources the field recognises less may find AI taking a larger role in meeting demands. The author states repeatedly that this is not an observed division among groups and requires empirical work on dispositions, enforcement outcomes and institutional context.

What this means for practice

• Governance offices and quality leads: Treat the recurring vocabulary as a signal to write the operational definitions yourselves. The paper's recommendation is disclosure rules that specify when AI use must be declared, how it is recorded, and what the institutional response is when it is not — plus review of whether sanction procedures fall unevenly across student groups.
• Assessment designers: The corpus's own move is toward process evidence, but it stops at recommendation. The paper argues for complementing detection with process-based evidence and oral examinations, staged drafts and required reflections that show how AI influenced the work, since textual traces cannot distinguish enhancement from substitution.
• Leadership in fast-adopting systems: The paper's practical claim for leaders is a shift from policy implementation to policy enactment, meaning active interpretation and translation of international principles into local expectations rather than passive application. It calls this critical governance literacy: the ability to tell an ethical aspiration from an actionable rule, and to notice when responsibility has quietly moved onto individual judgment.
• Researchers and equity officers: The capital-filter hypothesis is the part worth testing. It needs comparative evidence on patterns of AI use, detection outcomes and subsequent misconduct decisions; without it, the fairness of integrity processes rests on an assumption the paper argues is unexamined.

Limitations

• The corpus is four publications from two intergovernmental organizations, mixed in genre and institutional authority, and the author states they cannot be read as a unified policy position. National and institutional policy, where enforcement actually happens, is outside the analysis.
• The design is conceptual. The study provides evidence of how policy discourse constructs academic integrity, not evidence of student behavior, cognitive experience or dispositional change, and the author says so at each point where the habitus is invoked.
• The changing cheating habitus, its stratification by academic capital and the detection-as-capital-filter proposition are theoretical inferences from discourse. No comparative data on AI use, flags or sanction outcomes support them.
• The contextual literature comes from South Africa, the Gulf and work-integrated learning settings, and the author flags that these mechanisms are not established to operate the same way in the UAE, where the researcher's own institution sits.
• The researcher's dual position as both analyst of AI governance discourse and an educator inside an institution responding to it is disclosed, with reflexivity described as a mitigating practice rather than a removed vantage point.

Citation

Hashem, R. (2026). Cheating or being cheated? Academic integrity and a changing cheating habitus in international AI governance discourse. Frontiers in Education, 11, 1953305.