Facing 15-20% AI-cheating rates on some take-home work, one NZ school reverted to hand-written, supervised assessment; detection reportedly fell to ~5%. NZQA responded with a national policy banning GenAI in exams and warning that AI detectors "do not work" reliably.
15-20 %
AI-cheating rate detected at one school before format change (2023-2024)
~5 %
AI-cheating rate detected at the same school after reverting to hand-written, supervised assessment (2024)
19 standards
NCEA Level 1 standards where written reports dropped as external-assessment method from 2025 (from 2025)
Details
Maturity
Established
Promoter
New Zealand Qualifications Authority (NZQA) / Ministry of Education
Period
2023–ongoing (Level 1 assessment-method change from 2025)
Keywords
academic integrity, assessment policy, national qualifications, AI-detection scepticism, secondary education
Context
Since 2023, updated through 2025, the New Zealand Qualifications Authority (NZQA) and the Ministry of Education have issued national guidance prohibiting chatbots, generative AI and paraphrasing tools in NCEA external assessment, so candidates cannot submit AI-generated material as their own work. Schools holding 'consent to assess' must adopt their own authenticity policy for internal assessment, specifying acceptable AI use. The guidance explicitly warns that AI-detection software should not be relied on alone because of false positives, and instead points teachers toward authentication strategies grounded in knowing their students' own work.
Objectives
The policy aims to preserve the authenticity of NCEA assessment as generative AI tools become widespread, without over-relying on unreliable AI-detection software. It promotes concrete authentication strategies for teachers, including milestone check-ins, source referencing, follow-up questioning, comparison against a student's prior work, and watching for stylistic anomalies.
Activities
RNZ reporting in 2024 described individual schools — including Saint Patrick's College Wellington, Onslow College and Westlake Girls' High School — responding to detected AI cheating on take-home assessments by reverting to hand-written, supervised formats. Nationally, the Ministry and NZQA decided that written reports would no longer be used as the external-assessment method for 19 NCEA Level 1 standards from 2025, citing both administrative burden and authenticity concerns.
Results
At one school quoted in the RNZ investigation, 15-20% of students were found to have used AI to cheat on early take-home assessments; after the school reverted to hand-written and supervised formats, detected AI-assisted cheating reportedly fell to roughly 5%. Teachers described the shift as having made essay-based assessment 'really tricky', with one noting the school had 'reverted back to hand-writing for a lot of assessments'.
Conclusions
The practice's own description cautions that these figures come from individual schools quoted in a single news investigation rather than a controlled national study, and reflect detected or self-reported cheating, not a rigorous causal evaluation. It also notes that the shift toward hand-written and supervised assessment carries acknowledged classroom costs, and that no equity-specific provisions were found in the published guidance for students who rely on assistive or translation technology.
Implementation
Indicative cost
Low (< €50k)
Time to results
Long (> 3 years)
Staffing & skills
NZQA and Ministry of Education policy staff issuing and updating national guidance, School-level teachers implementing authentication strategies (milestone check-ins, source referencing, stylistic review), Schools holding 'consent to assess' status responsible for authoring and applying their own AI-authenticity policy
Conditions for success
Schools must hold 'consent to assess' and adopt a documented AI-authenticity policy for internal assessment
Teachers need to know their students' typical work well enough to spot stylistic anomalies rather than relying solely on AI-detection software
National guidance explicitly discourages reliance on AI detectors alone, given acknowledged false-positive risk
Common failure modes
Over-reliance on unreliable AI-detection software producing false positives
Increased teacher workload from hand-written/supervised assessment and follow-up questioning
No equity-specific provisions identified for students who rely on assistive or translation technology
Where it fits
Governance type
national
Scale
national (NCEA, all secondary schools)
Income level
high-income
Data sources
Where this practice's information was retrieved from, and when.
The Caribbean Examinations Council published binding, region-wide standards governing AI use in CSEC, CAPE and CCSLC school-based assessments across its …
The University of Latvia's Senate approved a binding AI-use regulation, effective 16 March 2026, built on integrity, transparency, responsibility and …