Weekly edition

Policy catches up, trust doesn't

2 July to 13 July 2026

The eleven days to 13 July were less about new platforms than about the widening gap between policy and practice. Four more US states turned AI guidance into law, while a Brown University class where a professor suspects most students used AI to cheat showed exactly why policy on paper doesn't stop misuse in the room. New UK research raised a quieter worry: AI quietly taking over the group work schools rely on to build collaboration skills. And an unresolved case from home, AI-generated deepfakes targeting teachers, is a reminder that trust in AI at school is about more than academic integrity.

Ngā tohu | Signals this week

  1. More US states are turning AI guidance into law: Idaho, Maryland, Oklahoma and Virginia all now require school boards to adopt their own AI policies aligned to state standards, with Oklahoma going as far as banning AI from grading or discipline decisions.
  2. A Brown University class where a professor suspects most students used AI to cheat on a take-home exam is the sharpest reminder yet that assessment design, not detection, is the only defence that scales.
  3. New UK research raises a quieter concern: as students turn to AI to summarise and smooth over group members' contributions, the interpersonal skills group work is meant to build risk being designed out.
  4. At home, June's still-unresolved case of AI-generated deepfakes targeting teachers is a reminder that trust in school AI is not only about cheating.

Policy and regulation

K-12 Dive9 July 2026

4 more states require districts to adopt AI policies

New laws in Idaho, Maryland, Oklahoma and Virginia require state education departments to issue AI guidance and require school boards to adopt their own aligned policies. Oklahoma's law goes further, banning AI from being the primary basis for grading, discipline or other high-stakes decisions, and letting parents opt students out without academic penalty.

He aha te take? Why it matters for AotearoaNew Zealand hasn't taken the legislative route these states are now on: our approach stays guidance-led, with the Ministry's design-not-detection stance as the operative document rather than statute. Oklahoma's law requires schools to annually disclose every AI tool in use and what student data it collects, which is a sharper transparency bar than most NZ schools currently meet. Boards reviewing their own AI policy this term could reasonably ask whether whānau get that same clarity, and whether it lines up with the Privacy Commissioner's IPP3A guidance.
Read the source →

Research and evidence

Inside Higher Ed8 July 2026

Brown professor suspects majority of his class used AI to cheat

A Brown University economics professor gave a take-home midterm for the first time and watched the class average jump to 96 percent. Suspecting mass AI use, he moved the final exam in-person; more than a dozen students dropped the course, and the class average on the final fell to a historic low of 48.6 percent.

He aha te take? Why it matters for AotearoaThis lands right in the middle of the assessment-design conversation NZ has been having since the Assessment Design Forum and the curriculum consultation work under way: it's a live example of what happens when a take-home, unlimited-time format meets generative AI at scale. NCEA's shift toward more internal assessment makes this worth reading closely. The Ministry's own guidance, design out misuse rather than detect it, is exactly the lesson Brown learned the hard way, and departments trialling take-home or open-book internal assessments should treat this as a stress test of their own design, not just a cautionary headline from overseas.
Read the source →
HEPI8 July 2026

Is AI quietly eroding the social core of student teamwork?

A UK higher education researcher argues that as students increasingly use AI to summarise peers' contributions or resolve group disagreements, the process of explaining, questioning and negotiating that makes group work educationally valuable is being quietly bypassed, even as final outputs look fine.

He aha te take? Why it matters for AotearoaCollaboration and oral language carry real weight in the New Zealand Curriculum and in kaupapa Māori approaches to collective sense-making, so this argument travels well here. The practical takeaway for kaiako is concrete: if AI can produce a polished group output with almost no real interaction between students, then the assessment needs to capture how the group worked, not just what it handed in. That might mean building in moments where groups have to explain, question and disagree with each other in front of you.
Read the source →

Aotearoa spotlight

RNZ22 June 2026

Students use AI to make pornographic images of teachers

Multiple NZ teachers and principals have been the target of AI-generated deepfake pornographic images and videos made by students, leaving some too distressed to work. Netsafe and legal experts say New Zealand, unlike Australia, has not banned the 'nudifying' apps and tools behind the harm, and the Deepfake Digital Harm and Exploitation Bill before Parliament would help but wouldn't address the scale of the problem.

He aha te take? Why it matters for AotearoaThis story broke just before this fortnight's window but belongs in this edition because it reframes what 'trust in school AI' actually means. The cheating and assessment stories above are about students using AI against the rules; this is AI being used to cause real, targeted harm to staff. Schools need a clear Harmful Digital Communications Act pathway and Netsafe contact ready before an incident lands, and RTLB and pastoral care staff should treat this as a live wellbeing risk, not a hypothetical one.
Read the source →

Published 22 June 2026, just before this edition's 2 to 13 July window. Featured in the spotlight because it directly frames this fortnight's assessment-integrity stories with a harder question about trust and safety.

What to do with this

School leadersA written AI policy is not the same as a tested one. Brown's experience shows what happens when a policy meets a real, large-scale incident: ask whether yours has ever been pressure-tested against a scenario like it, and whether staff know what happens next if it did.
KaiakoIf group work matters to you, make the process visible, not just the output. Build in moments where groups have to explain and defend their thinking to each other and to you, so AI can't quietly become the go-between.
RTLB and learning supportDeepfake misuse is a wellbeing and safety issue as much as a technology one. Know your school's Harmful Digital Communications Act pathway and your Netsafe contact before an incident lands, not after.
Boards and whānauAsk your school what its AI policy would actually do if it suspected mass AI-assisted cheating tomorrow. A policy that has never been tested against a real scenario is mostly a statement of intent.

Browse past editions →

On the radar at home

ERO report on AI in schools Still not published as at 13 July. Sector feedback was delivered in early July; publication remains imminent.
Senior secondary AI curriculum Year 11 to 13 AI content is scheduled to become mandatory between 2028 and 2030. Consultation timeline to watch.
Ministry guidance on generative AI The operative guidance on assessment authenticity: design not detection, and generative AI remains banned in external NCEA assessment.
Privacy Commissioner IPP3A guidance Final guidance (December 2025) is now the operative document for any tool collecting student information indirectly.
Deepfake Digital Harm and Exploitation Bill Before Parliament; would expand the definition of 'intimate visual recording' to cover AI-generated and altered images. Relevant to schools following this fortnight's Aotearoa spotlight story.