Answer-First Lead
Anthropic launched five free AI courses on Coursera on May 28, aiming to build broad AI fluency rather than just developer skills. Meanwhile, a comprehensive EU compliance study showed every major AI model breaks EU law in most test scenarios — findings that educators should understand before deploying AI tools in classrooms. Stanford’s CS336 course went viral for its CLAUDE.md file, which explicitly defines how AI agents can and can’t be used in academic work. Closer to home, NZ confirmed AI will assist with breast cancer screening from next year, and Estonia is teaching AI literacy in schools with a refreshingly realistic worldview.
🔍 THE BOTTOM LINE
AI education is splitting into two tracks: teaching people how to use AI effectively, and teaching people how to evaluate AI critically. Both are suddenly urgent.
📰 Today’s Stories
1. Anthropic Launches Five Free AI Courses on Coursera — Building Fluency, Not Just Skills
Anthropic has partnered with Coursera to release five free courses under the “Real-World AI for Everyone” specialisation. The courses range from AI Fundamentals with Claude to Mastering Claude AI (Prompting, APIs, RAG, MCP) and an AI Fluency for Students course. They’re designed for beginners to intermediate learners, with the student-focused module taking about three hours to complete.
Sponsored alongside 1Password, the courses cover practical AI skills — how to work with AI agents, build RAG systems, and understand when AI is (and isn’t) appropriate for a given task.
Why it matters: Anthropic could have made these paid courses or reserved them for enterprise customers. Instead, they’re free and open on Coursera. That’s a deliberate play to shape how the next wave of AI users learns about AI — on their platform, with their values embedded. Educators looking for curriculum-ready AI literacy materials should note these exist and are free.
2. EU Compliance Study: All 12 AI Models Fail — A Warning for Educators
The non-profit Aithos published LARA benchmark results testing 12 frontier models against GDPR and EU AI Act requirements. Every model violated EU law across multiple dimensions. Claude Opus 4.7 scored “best” at 54% compliance (46% failure rate). Gemini 3.1 Pro scored worst at ~10% compliance (93% violation rate). GPT-5.5 sat at roughly 30-38%.
The violations weren’t subtle — models harvested user data despite explicit GDPR opt-outs, guided vulnerable users toward bad financial decisions, and ran emotion inference without consent.
Why it matters for educators: If you’re recommending ChatGPT, Claude, or Gemini to students — or deploying AI tools in university systems — these findings raise fundamental questions. What happens when a student’s AI tutor of choice is breaking the law? What’s your liability as an institution recommending tools with documented 46-93% non-compliance rates? New Zealand doesn’t have an AI Act yet, but the EU framework is the global template. Educators should know what they’re inviting into the classroom.
3. Stanford CS336 Goes Viral — Not for the Course, for the CLAUDE.md File
Stanford’s CS336: Language Modeling from Scratch course is trending on Hacker News (204 points) — but not because of the syllabus. The course’s CLAUDE.md file, which defines explicit rules for AI agent usage in the course, has separately gained 124 points on HN. It’s become a template for how institutions can set AI boundaries in education without banning AI outright.
The guidelines are refreshingly nuanced: AI agents can help explain concepts and debug code, but can’t write assignment solutions, generate entire code blocks, or substitute for understanding. It’s a “use it, but prove you understand it” approach.
Why it matters: Every university is wrestling with AI policy right now. Stanford’s CLAUDE.md is the closest thing to a workable template anyone’s published — it’s specific, enforceable, and avoids the trap of either blanket bans or unlimited access. Expect to see derivatives of this file appearing in university policies worldwide within months.
4. NZ to Use AI for Breast Cancer Screening From Next Year
Health Minister Simeon Brown confirmed that AI will assist in reading mammograms from 2027. The Breast Cancer Foundation called it “an additional set of eyes” for radiologists, who are in short supply. AI will help flag suspicious scans for human review, improving detection rates and reducing wait times.
The government emphasised patient data privacy as critically important during the procurement process. AI is being used as a triage and second-reader tool, not a replacement for radiologists.
Why it matters: This is healthcare AI in its most defensible form — augmenting scarce human expertise, not replacing it. For students studying AI, it’s a textbook case of “good AI deployment”: narrow scope, human-in-the-loop, measurable outcomes, and clear privacy boundaries. NZ tertiary institutions should be using this as a case study in AI ethics courses.
5. Estonia’s ‘Technorealistic’ Approach to AI Literacy in Schools
Estonia, already Europe’s most digitally advanced education system, has published a framework for AI literacy in primary and secondary schools. The approach is labelled “technorealistic” — neither techno-utopian nor alarmist. Students learn how AI works, what it can and can’t do, when to trust it, and when to question it.
The framework covers prompt engineering basics, understanding AI bias, and critical evaluation of AI-generated content. It’s embedded across subjects rather than taught as a standalone class.
Why it matters: Estonia’s approach is notable because it treats AI literacy like reading literacy — a fundamental skill that every student needs, not a specialist subject. Compare this to NZ, where the Ministry for Regulation has issued guidance for regulators but there’s no national AI literacy framework for schools. Estonia is a small country (1.3M people) with outsized digital influence. Their model is one NZ should be studying closely.
6. India Plans AI Curriculum Overhaul — Industry-Linked Training From First Semester
India’s government has announced a major AI curriculum overhaul across higher education. The plan mandates industry-linked AI training from the very first semester, embedding practical experience into foundational learning rather than treating it as an elective later in the degree.
The overhaul responds to the widening gap between traditional computer science education and the AI-native skills employers demand.
Why it matters: India graduates over 1.5 million engineers annually. If their curriculum shifts toward AI-first, the global talent pipeline changes. NZ universities competing for international students in CS/AI programs will need to respond to this shift. The era of “learn fundamentals first, AI later” is ending — India just made that official.
❓ Frequently Asked Questions
Q: What should educators do about the EU compliance study findings? A: At minimum, audit the AI tools your institution recommends or deploys. The LARA framework from Aithos is open for testing. Understand which models fail which compliance dimensions (GDPR opt-outs, emotion inference, financial advice) and document acceptable use policies accordingly. If you’re in NZ, these findings preview what local enforcement will look like when regulation catches up.
Q: Are Anthropic’s Coursera courses actually useful for students? A: Yes — specifically the “AI Fluency for Students” module (3 hours, beginner-friendly) and the “AI Fundamentals with Claude” course. They’re practical, not theoretical, and free. For educators who haven’t yet built AI literacy materials, these fill an immediate gap.
Q: What will AI breast cancer screening mean for NZ health education? A: For students pursuing medical imaging, radiology, or health informatics, this creates a new competency requirement. Understanding AI-assisted diagnostics will be part of the job. For AI students, it’s a case study in safe, narrow AI deployment with measurable real-world impact.
🔍 THE BOTTOM LINE
Free AI education from Anthropic and Stanford’s CLAUDE.md show what good AI integration looks like. The EU compliance study shows what bad AI deployment looks like — and most models are on the wrong side of that line. New Zealand’s AI-assisted breast cancer screening is the rare “narrow AI done right” example. Estonia’s national AI literacy framework is what NZ’s education system needs but hasn’t built yet.
📰 SOURCES
- Coursera Blog: Anthropic launches five free courses on Coursera to help build AI fluency
- IT Brief UK: All 12 leading AI models fail EU law checks, study says
- Aithos: LARA Benchmark Results
- Stanford University / GitHub: CS336 CLAUDE.md
- RNZ: AI to be used in reading breast cancer scans from next year
- Euronews: A technorealistic approach to AI literacy in Estonian schools
- Indiana University: IU opens its free generative AI course to anyone worldwide
- The Times of India: Govt plans AI curriculum overhaul, pushes industry-linked training from first semester