NZ Issues AI Guidance for 267 Regulators — While Copyright Suits Threaten the Foundations
New Zealand’s Ministry for Regulation released Responsible AI in Action guidance for the country’s 267 regulatory bodies, with regulation minister David Seymour saying it will help them “do more, and do it faster.” The guidance recommends AI for low-risk tasks like triaging and data validation, with humans remaining in the loop for judgement and accountability. It warns that AI won’t fix a weak system — it’ll make problems “bigger, faster, and harder to explain.”
But the same day, RNZ reported on the copyright risks of NZ’s AI push. Copyright Licensing NZ’s Sam Irvine warned the government risks “paying subscriptions to international companies who are selling products that are part of international lawsuits.” Five major US publishers just sued Meta over AI training on copyrighted works. OpenAI lost a copyright case in Germany. Victoria University law professor Graeme Austin said the rush was “fraught” given the unresolved global litigation.
Why it matters for education: Schools and universities are deploying AI tools built on the same contested foundations. If courts rule against AI companies, what happens to the ed-tech products already integrated into curricula? NZ’s guidance is a start, but it ducks the copyright question entirely.
Pope Leo’s Encyclical: AI Must Serve Humanity, Not Concentrate Power
Pope Leo XIV’s Magnifica Humanitas — the first papal encyclical on AI — is a direct intervention in the AI governance debate. The encyclical argues AI must be “disarmed” and that the concentration of AI power in a handful of US tech firms threatens human dignity. It calls for epistemic humility — the recognition that not all knowledge should be automated.
For educators, the most relevant passage is the Pope’s insistence that AI should augment human judgement, not replace it — mirroring the exact language in NZ’s regulatory guidance. The encyclical is being read in Catholic schools worldwide, and its framing of AI as a moral question, not just a technical one, is already influencing curricula.
Why it matters for education: When the Vatican and the NZ Ministry for Regulation use the same language about human oversight, something is shifting in the AI governance conversation. Educators need to engage with these moral frameworks, not just the technical ones.
DeepSWE: The Benchmark We Use to Grade AI Coding Is Wrong a Third of the Time
Datacurve’s DeepSWE benchmark found that SWE-Bench Pro — the most widely used AI coding evaluation — has verifiers that issue incorrect verdicts 32.5% of the time. They accept wrong solutions 8.5% of the time and reject correct ones 24% of the time.
This matters for education because these benchmarks are increasingly used to assess AI coding tutors, AI-assisted learning tools, and even to evaluate student progress in AI-augmented coding courses. If the grading system is broken by a third, what are we actually measuring?
Why it matters for education: Every coding bootcamp, university CS department, and ed-tech platform that relies on SWE-Bench scores to evaluate AI tools is working with flawed data. The benchmark crisis is an education crisis in disguise.
Box CEO’s ‘AI Psychosis’ Warning: What It Means for Education
Aaron Levie’s diagnosis of “AI psychosis” in CEOs — the tendency to conflate AI prototypes with production-ready capability — applies directly to education. When school administrators see AI grade papers, generate lesson plans, or tutor students in demos, they make the same leap: “It works here, so it must work everywhere.”
But the data tells a different story. A UC Berkeley meta-analysis found “no robust relationship between AI adoption and aggregate productivity gain.” MIT researchers predict AI will reach “minimally sufficient quality” on most text tasks by 2029. We’re three years away from baseline competence on most tasks — and we’re deploying AI in classrooms now.
Why it matters for education: Education leaders need the same reality check Levie is giving CEOs. A demo of an AI tutor is not a working AI tutor. A grading prototype is not a reliable grading system. The gap between what AI can demo and what it can deliver reliably is where educational harm happens.
🔍 THE BOTTOM LINE: Three stories converge today: NZ issues AI guidance that says humans must stay in the loop, the Vatican says AI must serve humanity not concentrate power, and the benchmark we use to measure AI ability is broken by a third. The message for education is clear — AI tools in the classroom need the same scrutiny we’d apply to any other assessment instrument. If the grading system is wrong, nothing built on it can be trusted. The humans-in-the-loop principle applies as much to education as it does to regulation.