Australian universities split over using AI to mark student work as institutions adopt divergent policies

Divergent institutional policies

A clear split has emerged among Australian higher‑education providers regarding the use of generative artificial intelligence for assessing student submissions. The University of New South Wales, the University of Melbourne and the University of Sydney have issued blanket prohibitions that prevent teaching staff from employing AI tools to evaluate coursework. In contrast, the University of Adelaide, the University of Newcastle, Deakin University and RMIT have adopted a conditional approach that allows staff to use AI‑assisted marking provided the academic retains final responsibility for the grade and feedback.

These contrasting stances reflect differing risk assessments and strategic priorities. The prohibitive group cites concerns over academic integrity and the reliability of machine‑generated evaluations, whereas the permissive institutions emphasize potential efficiency gains and the possibility of AI serving as an auxiliary reviewer.

Conditional use and opt‑out mechanisms

Where AI assistance is permitted, universities have introduced safeguards. The University of Newcastle, for example, offers students the right to opt out of AI‑assisted marking, ensuring that their work is evaluated solely by human assessors. Deakin University explicitly states that staff may use AI to improve efficiency and support assessment activities, but the system may not assign grades autonomously. Western Sydney University reinforces a "human‑centred" model, insisting that the awarding of marks, grades and feedback always remains the responsibility of academic staff.

These policies aim to balance the desire for workload relief with the need to preserve academic standards and student trust. By retaining final human judgement, institutions attempt to mitigate the risk of erroneous or biased AI outputs while still exploring the technology's supportive role.

Institutions still in the exploratory phase

Several universities have not yet deployed AI for marking but are integrating the technology into broader teaching and assessment practices. Monash University, the University of Wollongong and James Cook University report that AI is not used in the marking process yet, yet it is being incorporated to support student learning through tools such as writing assistants and personalised feedback generators. The University of Queensland is developing policies that explicitly ensure academics continue to assess work directly, signalling a cautious approach to automation.

These exploratory steps illustrate a tentative adoption curve, where institutions pilot AI in low‑stakes contexts before considering any expansion into high‑impact grading decisions.

Critics warn of a potential "slop‑cycle"

Opponents of widespread AI marking argue that it could trigger a "slop‑cycle" in which students increasingly submit AI‑generated assignments that are subsequently graded by AI. This feedback loop may lead to verification drift, where the authenticity of work becomes harder to confirm, and hallucinated content could slip through unchecked. Critics contend that such a cycle could erode university reputations if low‑quality or fabricated submissions are accepted as legitimate academic output.

The warning highlights a systemic risk: once AI becomes a primary evaluator, the incentive for students to rely on AI generation may intensify, potentially diminishing the development of critical thinking and writing skills.

Workload pressures and the promise of a "second pair of eyes"

Academic staff across Australia report being overworked, with marking consuming a substantial share of their limited time. Some lecturers view AI as a prospective "second pair of eyes" capable of flagging inconsistencies or plagiarism, thereby easing the manual burden. However, the benefits remain largely untested, and staff members have voiced concerns that reliance on AI could lead to a "disaster" if the technology fails to capture nuanced judgement required for complex assignments.

The tension between potential efficiency and unproven reliability underscores the need for rigorous evaluation before scaling AI‑assisted marking.

QUT’s assessment uplift project and staff concerns

Queensland University of Technology (QUT) has launched an "assessment uplift project" designed to keep assessment authentic, trusted, scalable and future‑focused in an AI‑enabled environment. While the initiative seeks to modernise evaluation practices, staff have expressed apprehension that the project could reduce the volume of human marking required, thereby threatening the primary source of income for sessional academics who depend on marking workloads.

The QUT case illustrates how institutional strategies to embed AI can intersect with labour considerations, raising questions about the future role of contract teaching staff.

Student resistance to AI‑driven assessment

Student sentiment adds another layer of complexity. A QUT creative writing student, Alex Cameron, articulated a strong objection to AI grading, stating that he would not pay for a degree if marking were performed by machines. Cameron’s comment reflects broader student resistance to AI‑driven assessment, particularly in disciplines where nuanced feedback is valued.

Such resistance suggests that any move toward automated grading must consider student expectations for human interaction and mentorship.

Uncertain effectiveness and lack of empirical data

University representatives acknowledge that the effectiveness of AI‑assisted marking at scale remains unknown. Claims of improved quality or efficiency are largely untested, and no empirical data on outcomes or reliability has been presented. This admission leaves the impact of AI on assessment open to further evaluation and underscores the importance of systematic research before institutional roll‑out.

Without robust evidence, policy decisions risk being driven by speculation rather than demonstrable benefit.

Practical implications for organisations

For organisations that partner with or employ graduates from Australian universities, the current landscape signals a need for vigilance. Employers should be aware that assessment standards may vary widely depending on the institution’s AI policy, potentially affecting the consistency of graduate competencies. Where AI marking is permitted, organisations might consider supplementing credential verification with additional skill assessments to ensure that graduates have received rigorous human evaluation. Conversely, institutions that prohibit AI grading may produce graduates with more traditional feedback experiences, which could be advantageous for roles requiring nuanced analytical abilities. In all cases, transparent communication with educational partners about assessment practices can help align expectations and mitigate risks associated with divergent grading approaches.

• University of New South Wales, University of Melbourne, University of Sydney – prohibit AI marking.
• University of Adelaide, University of Newcastle, Deakin University, RMIT – allow AI assistance with human oversight.
• Queensland University of Technology – assessment uplift project aiming for AI‑enabled authenticity.
• Monash University, University of Wollongong, James Cook University – integrating AI in teaching but not marking.