Academic Integrity Stress-Test

Workbook-grounded and author-clarified; proposed reconstruction. Before an assessment is released, educators need to review how its design, evidence, instructions, permitted assistance, policy, marking and support operate together. Current GenAI, collaboration and online resources create both legitimate uses and vulnerabilities, while proposed controls may create false-positive, access and privacy harms.

Proposed and author-clarified. Integrity review often occurs after release or focuses narrowly on detecting prohibited output. This encourages reactive policing and may leave weak learning evidence, ambiguous rules, policy conflicts, unrealistic marking, inaccessible controls and false-positive risks unexamined. Tool capability can be mistaken for evidence about a student.

Proposed and author-clarified. Conduct a comprehensive pre-release stress test of the assessment’s outcomes, evidence, instructions, criteria, permitted assistance, policy alignment, marking, moderation, accessibility and feasibility. Use bounded GenAI trials with synthetic or public inputs and compare plausible misuse with equally serious ambiguity and false-positive scenarios. Trace each finding to a proportionate revision of task, evidence, communication, support or judgement. Re-test and require accountable approval. Keep individual misconduct detection and adjudication outside the pattern.

• A documented threat and false-positive model for the assessment.
• Identified vulnerabilities in task design, evidence and communication.
• Proportionate revisions that strengthen learning evidence and clarify permitted assistance.
• An integrity review record that preserves due process and assessor authority.
• Bounded GenAI trial records that document assessment capability risks without becoming evidence about students.
• A teacher- and governance-approved release version with consistent student and marker guidance.

Educators and assessment designers own validity, evidence, learning and release approval. Integrity and policy specialists advise on rules and process; accessibility, privacy, moderation and student perspectives challenge harms and ambiguity. GenAI systems produce bounded trial outputs and scenarios. They cannot determine misconduct, identify an offender, establish proof or recommend sanction.

Reconstructed risks

• The exercise generates actionable cheating strategies without adequate containment.
• Unreliable detection claims are embedded in policy or marking decisions.
• Revisions increase surveillance, workload or inequity more than they improve validity.
• Legitimate accessibility or language support is mistaken for misconduct.
• A technical adversarial mindset displaces educational purpose and student trust.
• Trial outputs or prompts are circulated in ways that create a reusable evasion guide.
• A pre-release design finding is later misrepresented as evidence in an individual misconduct case.

• Minimum viable implementation: review the complete draft against a short integrity and fairness checklist, run several bounded GenAI trials with public or synthetic inputs and revise the task before teacher approval.
• Robust implementation: add a versioned assessment contract, recorded trial conditions, vulnerability and false-positive register, policy and accessibility review, traceable revisions, re-testing and formal release approval.
• Low-code implementation: connect a controlled assessment store, approved GenAI trial workspace and structured review register while preventing trial findings from entering student case systems.
• Local or privacy-preserving implementation: keep unreleased assessments and sensitive trial details in controlled storage, use no student data and process locally where exposure or policy risk warrants it.
• Speculative implementation: maintain a reviewed suite of synthetic stress tests for recurring assessment types and flag designs for renewed review when tools or rules change, without automating release decisions.