The regulatory and pedagogical evidence behind LogAI.

“Methods of assessment are consistent with the learning outcomes being assessed, are capable of confirming that all specified learning outcomes are achieved and that grades awarded reflect the level of student attainment.”

Higher Education Standards Framework (Threshold Standards) 2021 (Cth) sch 1 s 1.4(3).

Summary

HESF requires providers to do more than state learning outcomes. They must use assessment methods that can confirm those outcomes have actually been achieved, and they must also maintain academic integrity through preventative action, misconduct processes, and student guidance. In the GenAI context, the implication is that institutions need auditable assessment governance and integrity processes, not detector outputs or unsupported suspicion.

“it appears almost impossible to detect if these technologies have been used in the production of assessment products in a reliable way.”

(TEQSA, 2023, p. 2)

Summary

TEQSA's 2023 discussion paper pivots the sector from detection-based to redesign-based assessment integrity. The current canonical frame (per the September 2025 revised PDF) is the 'two lanes' architecture: a secure lane producing direct attainment evidence under HESF 1.4.3 / 1.4.4, and an AI-permitted lane with disclosure and process evidence.

“it is unwise to rely solely on AI detectors as a means of managing the risk to academic integrity posed by AI.”

(TEQSA, 2024, p. 3)

Summary

TEQSA's 2024 update names AI detection as an inadequate standalone response and directs institutions toward structural redesign with transparent, auditable process evidence.

“Rather than investing primarily in detection mechanisms, institutions need to emphasise the redesign of assessment to capture authentic demonstrations of student capability and comprehension.”

(TEQSA, 2025, p. 4)

Summary

TEQSA's September 2025 revised implementation PDF introduces the 'two lanes' architecture used sector-wide: a secure lane producing direct attainment evidence under HESF 1.4.3 / 1.4.4, and an AI-permitted lane with disclosure and process evidence. Process evidence and authentic engagement are framed as the throughline.

“The knowledge hub provides links to a broad range of resources from TEQSA, other government agencies, across the sector and international organisations.”

Tertiary Education Quality and Standards Agency (TEQSA). (2025, May 5). Updated Gen AI knowledge hub now on our website.

Summary

This TEQSA knowledge-hub update is a curation source rather than a standalone regulatory standard. It shows TEQSA consolidating academic-integrity, research, student-support, and government resources into a single institutional reference point, which signals that providers are expected to keep pace with a broad and evolving evidence base.

“The toolkit includes 3 dimensions: Process, People and Practice. Each dimension can be read individually, in any order, or in conjunction with the broader toolkit.”

Tertiary Education Quality and Standards Agency (TEQSA). (2024). Gen AI strategies for Australian higher education: Emerging practice.

Summary

TEQSA's 2024 toolkit translates the sector-wide GenAI request for information into practical patterns across process, people, and practice. Its emphasis is not on one universal assessment fix; it is on institutions building documented, context-specific responses that protect award integrity while still integrating AI ethically into teaching and learning.

“The Framework has a central focus on equity to align with the Australian Universities Accord Final Report recommendations (Department of Education, 2024) and to avoid amplifying existing digital divides and social inequities (Birhane et al., 2022).”

Australian Centre for Student Equity and Success (ACSES). (2025). Australian Framework for Artificial Intelligence in Higher Education (p. 3).

Summary

ACSES frames AI governance in higher education through an explicit equity lens. The framework says AI implementation should avoid worsening existing digital divides and social inequities, and it aligns that work to the HESF and broader institutional governance obligations rather than treating AI as a standalone technology issue.

“Students should acknowledge the use of generative AI language models in assessment tasks, following any guidelines provided by their institution.”

Australian Academic Integrity Network. (2023). AAIN Generative Artificial Intelligence Guidelines (p. 2).

Summary

AAIN frames generative AI as an academic-integrity and learning-design issue that requires explicit acknowledgement, clear student guidance, policy refresh, and AI literacy. The guidelines connect integrity management to fairness, communication, and institutional responsibility rather than detector-only enforcement.

“Taking a long-term developmental approach avoids the immediate need to deploy unreliable detection software, retains the usefulness of written assessments and nurtures academic writing skills.”

Quality Assurance Agency for Higher Education. (2023). Reconsidering assessment for the ChatGPT era: QAA advice on developing sustainable assessment strategies (p. 4).

Summary

QAA argues that sustainable assessment in a GenAI environment depends on assessment redesign, developmental writing support, authentic tasks, and reduced reliance on unreliable detection software. Its position is that institutions need long-term assessment strategy changes rather than reactive policing.

“Universities will adapt teaching and assessment to incorporate the ethical use of generative AI and support equal access.”

Russell Group. (2023). Principles on the use of generative AI tools in education. Principle 3.

Summary

The Russell Group principles frame institutional AI readiness around literacy, staff guidance, ethical adaptation of teaching and assessment, equal access, and transparent academic integrity. They explicitly connect integrity to clear policy and open discussion rather than fear-driven enforcement.

“Ofqual wrote to all awarding organisations in September 2023 to confirm that the use of AI as a sole marker of students’ work does not comply with our regulations.”

Ofqual. (2024). Ofqual's approach to regulating the use of artificial intelligence in the qualifications sector. Using AI to mark students' work.

Summary

Ofqual's policy paper sets out a precautionary regulatory approach for high-stakes qualifications. The key point for institutions is that AI cannot be the sole marker of student work, because high-stakes marking still requires human judgement, transparency, and protection against bias and unfairness.

“Each candidate (candidate being defined as someone for whom an entry is in place for the unit or qualification) must sign a declaration when submitting their coursework to their teacher for final assessment. Electronic signatures are acceptable. This is to confirm that the work is their own and that any assistance given and/or sources used have been acknowledged.”

Joint Council for Qualifications (JCQ). (2025). AI Use in Assessments: Your role in protecting the integrity of qualifications (section 7.1).

Summary

JCQ operationalises AI disclosure as part of normal assessment authentication. Candidates must acknowledge assistance and sources, and teachers must authenticate that the submitted work is the student's own, which makes disclosure plus human verification the baseline control.

“Kura and schools, with consent to assess DASS-listed standards, must have an authenticity policy to support assessment practice. This policy must include the acceptable use of AI.”

New Zealand Qualifications Authority (NZQA). (2024). Guidance on the acceptable use of Artificial Intelligence. Advice for kura and schools with consent to assess.

Summary

NZQA's guidance frames AI use through authenticity policy rather than detector-first enforcement. It expects schools to define acceptable AI use and recommends milestones, observation, self-reflection, authenticity declarations, and careful caution around false positives from AI detectors.

“Rather than banning AI, we can reimagine assessment to focus on what AI can’t easily replicate — critical thinking in context, lived experience, cultural reasoning, and human judgement.”

(Ako Aotearoa, 2024)

Summary

Ako Aotearoa recommends process-focused design, reflective layers, and oral/live elements — explicitly warning against over-reliance on detection tools.

“The Guidance stresses the need for educational institutions to validate GenAI systems on their ethical and pedagogical appropriateness for education.”

UNESCO. (2023). Guidance for generative AI in education and research (p. 2).

Summary

UNESCO's guidance frames GenAI governance around human agency, inclusion, equity, and validation. It explicitly says educational institutions need to assess whether GenAI systems are ethically and pedagogically appropriate for education rather than assuming that consumer AI tools are suitable by default.

“Countries should develop and disseminate clear guidelines for the use of generative AI in education.”

OECD. (2023). OECD Digital Education Outlook 2023: Towards an Effective Digital Education Ecosystem (p. 273).

Summary

OECD recommends clear guidance, adaptive governance, and stakeholder literacy rather than rigid restriction as countries respond to generative AI in education. The report treats AI adoption as a governance problem involving bias, privacy, data security, and accountability, not just a classroom tool problem.

“Put simply, the use of AI in learning evaluation does not conflict with accreditation standards, policies, or practices. Accreditation is never a reason to not implement technology solutions that leverage AI for learning evaluation. Since innovating to advance student success is a central tenet of accreditation expectations, C-RAC supports the exploration and application of transparent, accountable, and unbiased AI solutions within the practice of learning evaluation and credit transfer.”

Council of Regional Accrediting Commissions. (2025). Statement on the Use of Artificial Intelligence (AI) to Advance Learning Evaluation and Recognition (p. 4).

Summary

C-RAC's 2025 statement is about AI in learning evaluation, credit transfer, and prior-learning recognition rather than classroom assessment integrity. Its main contribution is to clarify that accreditors do not prohibit transparent, accountable, and unbiased AI in those transfer-related evaluation workflows.

“AI systems intended to be used to evaluate learning outcomes, including when those outcomes are used to steer the learning process of natural persons in educational and vocational training institutions at all levels;”

Regulation (EU) 2024/1689 of the European Parliament and of the Council of 13 June 2024 laying down harmonised rules on artificial intelligence (Artificial Intelligence Act) [2024] OJ L 2024/1689 annex III, section 3(b).

Summary

The AI Act classifies automated assessment scoring as high-risk — requiring risk management, data governance, documentation, record-keeping, and human oversight. Disclosure + reflection tools that keep decisions with humans are not high-risk.

“High-risk AI systems shall technically allow for the automatic recording of events (logs) over the lifetime of the system.”

Regulation (EU) 2024/1689 of the European Parliament and of the Council of 13 June 2024 laying down harmonised rules on artificial intelligence (Artificial Intelligence Act) [2024] OJ L 2024/1689 art 12(1).

Summary

Article 12 of the EU AI Act mandates automatic event-logging for high-risk AI systems (enforceable from August 2026). Required log content includes period of each use, reference database, input data for a match, and identification of natural persons involved in verification.

“Providers shall ensure that AI systems intended to interact directly with natural persons are designed and developed in such a way that the natural persons concerned are informed that they are interacting with an AI system”

Regulation (EU) 2024/1689 of the European Parliament and of the Council of 13 June 2024 laying down harmonised rules on artificial intelligence (Artificial Intelligence Act) [2024] OJ L 2024/1689 art 50(1).

Summary

Article 50 establishes the transparency obligation: providers must ensure users know when an AI is involved in the decision or interaction affecting them. The obligation is structural, not discretionary.

“The guidelines are intended as a resource to foster reflection, transparency, and responsibility in the use of AI, rather than as binding requirements.”

ENQA (European Association for Quality Assurance in Higher Education). (2025). Guidelines for the Responsible Use of AI in External QA (p. 1).

Summary

ENQA sets out governance expectations for AI in external quality assurance: reflection, transparency, responsibility, data protection, equity, and human oversight are baseline requirements. While the guidance is written for QA agencies, it signals that institutional AI evidence needs to be reviewable, explainable, and governed.

“Develop an institutional ethos and culture which emphasises integrity, honesty, trust, and respect, rather than let the conversations be dominated by suspicion, surveillance, and distrust.”

National Academic Integrity Network. (2023). Generative Artificial Intelligence: Guidelines for Educators (p. 10).

Summary

The NAIN guidelines ask institutions to review assessment validity, balance process and product, refresh policies, and build an integrity culture based on dialogue rather than surveillance. The guidance is notable for explicitly warning against suspicion-led responses and for pushing institutions to communicate permitted and non-permitted AI use clearly.

Summary

BJET 2025 empirical study across a mixed sample of authentic-task assessments at two UK universities found graders could not distinguish GenAI-assisted work even in high-authenticity tasks. The conclusion is that authenticity alone is insufficient — process visibility is the mechanism that works.

Summary

Corbin, Dawson & Liu (2025) coin 'enforcement illusion': discursive frameworks (declarations, traffic-light AI-use scales) lack the structural teeth of the systems whose language they borrow. At least one structural lane per course (sampled oral, in-class check, live demo) is required for the research case to close.

Summary

Corbin et al. (2025) empirically document how staff and students interpret AI acceptable-use lines differently. Ambiguity at the policy level predicts non-disclosure and non-compliance, which strengthens Gonsalves (2024) with a mechanism.

Summary

Daffern et al. 2023 is the peer-reviewed meta-analytic anchor for oral-verification design: structure, station count, and duration are load-bearing. Unstructured oral checks are below the reliability floor for any assessment use.

Summary

BJET 2025, N=68 Educational Technology sophomores at one Chinese university. Metacognitive support improved SRL task-strategy and self-evaluation behaviours; between-group achievement differences were not statistically significant. Reflection is the strongest currently-available form of non-automated learning evidence; it is not a classifier of AI use.

Summary

Neshaei et al. 2025 empirically show LLM support can improve the quality of reflective writing but also carries cognitive off-loading risk. The fidelity of reflection degrades with temporal distance from the AI interaction.

Summary

Khlaif et al. 2024 is the direct empirical check on the 'reflection is non-fakeable' claim: reflective essays can be AI-mimicked at non-trivial fidelity, which qualifies the claim to 'reflection is the strongest among imperfect options.'

Summary

The AICAI (Academic Integrity and Creativity in the Age of AI) model reframes GenAI from a threat to be banned to a capability students must account for: the locus of integrity shifts from the product to the reasoning.

Summary

Drawing on Ajzen's Theory of Planned Behaviour, Bandura's self-efficacy theory, and situational crime prevention, Leaton Gray et al. argue detection-first policy erodes the ethical pedagogy (transparency and student agency) that underpins sustainable integrity. Ethical pedagogies built on transparency and student agency outperform surveillance.

Summary

OpenAI's own AI-text classifier, launched January 2023, was withdrawn in July 2023 after correctly identifying only 26% of AI-written text and incorrectly flagging 9% of human writing.

Summary

Peer-reviewed primary study (2023) tested 14 AI-detection tools; none reached 80% accuracy and only five scored above 70%. Systematic bias toward classifying AI output as human-written; all tested tools collapsed under simple paraphrase attacks.

Summary

Guo et al. 2026 tested 13 detectors (7 open-source, 6 commercial) on ~280,000 paired human/AI student-assignment instances. Hybrid editing defeats detection 88% of the time. The 2026 anchor for the adversarial-robustness case and the strongest available evidence that detection is not a reliable integrity layer.

Summary

Gonsalves (2024) identifies fear of academic penalty, ambiguous guidelines, and inconsistent enforcement as the drivers of non-disclosure. Low-stakes frameworks consistently show higher honest-reporting rates.

Summary

Beyond Detection (2025) replicates the Gonsalves (2024) finding in an economics UG cohort, identifying an 'early opt-out' pattern — strategic non-disclosure driven by policy ambiguity and penalty anxiety.

“AI use is now almost universal. Some 95% of students report using AI in at least one way and 94% say they use generative AI to help with assessed work.”

Higher Education Policy Institute (HEPI). (2026). Student Generative AI Survey 2026 (executive summary, p. 4).

Summary

HEPI's 2026 survey shows AI use is now near-universal among UK undergraduates and already deeply embedded in assessed work. It also shows institutional support remains inconsistent, with many students still reporting uneven encouragement, changed assessment conditions, and concern about misconduct accusations.

“Clear guidance is urgently needed: students are calling for consistent, course-specific policies on AI use, combined with practical, ethical, and course specific education, support or guidance”

Jisc. (2025). Student perceptions of AI 2025. Key messages.

Summary

Jisc's 2025 report shows students want course-specific clarity, practical support, and equitable access rather than generic institutional statements about AI. The findings also connect AI confusion to wider concerns about employability, misinformation, deepfakes, data use, and uneven access to premium tools.

“ALTL involves understanding AI fundamentals, critically evaluating AI applications, and maintaining vigilance against misuse and bias.”

EDUCAUSE. (2024). AI Literacy in Teaching and Learning: Executive Summary.

Summary

EDUCAUSE treats AI literacy as an institution-wide capability spanning technical understanding, evaluative skills, practical application, and ethical considerations. The framework is aimed at students, faculty, and staff, which makes AI readiness a governance and curriculum issue rather than a narrow conduct issue.

Summary

Digital Education Council's 2024 global survey (n=3,839 across 16 countries) finds high student adoption coupled with widespread institutional inadequacy — adoption is outpacing governance. The 86% / 80% headline figures are widely reported in press but should be treated as 'reported' pending direct verification against the full PDF.

Summary

Stanford's peer-reviewed empirical study (2023) found all seven major AI detectors misclassified non-native English writing at ≥61% false-positive rate while scoring near-perfect on native writing. When TOEFL vocabulary was enriched, false positives dropped to 11.6% — proving the bias is linguistic, not behavioural.

“There is a growing digital divide in AI use, with male students, students on STEM and Health courses and more socioeconomically advantaged students more likely to use AI than others.”

Higher Education Policy Institute (HEPI). (2025). Student Generative AI Survey 2025. Key findings.

Summary

HEPI's 2025 survey shows both rapid mainstreaming of student AI use and a measurable participation divide. The report matters because it combines very high uptake with evidence that AI use, confidence, and support remain uneven across gender, discipline, and socioeconomic lines.

“The new framework embodies UNESCO's mandate by anchoring its vision of AI and education in principles of human rights, inclusion and equity.”

Miao, F., Shiohira, K., & Lao, N. (2024). AI competency framework for students (p. 6).

Summary

UNESCO positions student AI competence as a human-rights and inclusion issue, not just a technical skills issue. The framework expects students to develop ethical judgement, critical understanding, and human-centred design awareness alongside technical knowledge of AI.

“Academic misconduct policies can disadvantage neurodivergent students through ableist assessment design, surveillance technologies, and pedagogies of forced disclosure.”

Eaton, S. E. (2025). Neurodiversity and academic integrity: toward epistemic plurality in a postplagiarism era. Abstract.

Summary

Eaton's 2025 narrative review of 15 sources argues that academic-integrity frameworks can disadvantage neurodivergent students through ableist assessment design, surveillance technologies, and pedagogies of forced disclosure. The paper reframes academic integrity through epistemic plurality rather than compliance-first models.

“Concerns were raised around changes to ‘AI-proof’ assessments which may result in less accessible assessments for some.”

Fletcher, R. (2025). Technological solutionism: disability, higher education and generative artificial intelligence. Technology and consistency in higher education section.

Summary

Fletcher's 2025 paper argues higher education risks technological solutionism in its GenAI response. It notes that disabled students are still inconsistently supported, are rarely engaged directly in GenAI policy conversations, and may be harmed when 'AI-proof' redesign reduces accessibility.

“The main concerns are the inaccuracy of answers, the risk of academic integrity breaches, and the digital divide created by subscription costs.”

Zhao, X., Cox, A., & Chen, X. (2025). The use of generative AI by students with disabilities in higher education. Highlights.

Summary

Zhao, Cox, and Chen surveyed 124 students with disabilities and found that chatbots, rewriting tools, and translation tools are being used to overcome academic-writing barriers. The main concerns were inaccurate outputs, academic-integrity risks, and subscription-cost barriers, alongside a strong desire for policy input and training.

“More than two-thirds (69.1 per cent) of 18- to 34-year-olds recently used one of the many available generative AI tools, compared with less than 1 in 6 (15.5 per cent) 65- to 74-year-olds.”

Hegarty, K., McCosker, A., Kennedy, J., Thomas, J., & Parkinson, S. (2025, November 5). Australia is facing an AI divide.

Summary

This national Australian survey of more than 5,500 adults found that generative AI use is already stratified by age, education, occupation, remoteness, and language background. The article frames that pattern as an emerging AI divide rather than a neutral uptake curve.

“5.1 The school uses well-documented assurance of learning (AoL) processes that include direct and indirect measures for ensuring the quality of all degree programs that are deemed in scope for accreditation purposes. The results of the school's AoL work leads to curricular and process improvements.”

AACSB International. (2025). 2020 Guiding Principles and Standards for Business Accreditation (Standard 5.1, p. 43).

Summary

AACSB Standard 5 requires documented assurance-of-learning processes, direct and indirect measures, and evidence that results lead to curricular and process improvement. In a GenAI environment, that makes the credibility and documentation of assessment evidence central to AoL practice rather than peripheral.

“Schools must establish governance and policies for AI use, including these cases. They should supplement rather than replace the human element of accreditation by easing data gathering and manual tasks.”

AACSB International. (2025). AI Use Case Hub for Accreditation. Responsible section.

Summary

AACSB's AI use-case hub frames AI as a support tool for accreditation work, not a substitute for human judgement. The consistent message is governance first: schools can use AI to reduce documentation burden, but they still need policy, privacy, and verification controls around any AI-assisted process.

“Embracing smarter AoL isn't about appeasing accreditors; it's about future-proofing a school's quality and impact.”

AACSB International. (2025, October 8). Rethinking AoL-Moving From Compliance to Strategy.

Summary

AACSB's October 8, 2025 article argues that assurance of learning becomes valuable when it drives visible curriculum improvement rather than producing compliance paperwork. In a GenAI environment, that reframes AoL as an institutional capability for evidence, interpretation, and change.