Hey {{ first_name | human }},

This week I’ve been looking at what happens when AI meets the everyday business of school: learning maths, protecting pupil data and working out what children actually know. There’s plenty to discuss, and a little light relief at the end.

⚡️ Maths tutoring: Google reports encouraging classroom results, with human checks throughout. Independent test gains remain unproven.

🛡️ Privacy: Microsoft’s September agreement with US teachers’ unions reaches its 1 November implementation date.

📝 Assessment: Teachers are bringing more writing and discussion into class to see pupils’ thinking.

🇭🇰 Curriculum: Hong Kong proposes AI within a new primary technology subject, starting in 2028/29.

🔍 Wider AI: Claude’s invisible watermark is reportedly coming to more models. It won’t settle every authorship question.

⚡️ Google puts AI maths tutoring to the test

What it is: On 29 October, Google reported a 10-week UK trial involving over 1,100 pupils aged 12–15 on Eedi. Pupils received static hints, human tutoring or Gemini-supported tutoring. Human experts checked every AI response before it reached a pupil. AI-supported pupils performed similarly to those with human tutors on platform tasks and intermediate quizzes, but the gains did not appear on independent STAR tests.

Why this matters: For me, the distinction is between doing better with support and learning that transfers beyond it. Google suggests pupils covered fewer questions because tutoring took longer; that explanation still needs testing. These results support further research into supervised tutoring, rather than a claim that a general chatbot can replace a teacher.

🛡️ Microsoft’s school AI privacy agreement takes effect

What it is: Announced on 9 September, Microsoft’s agreement with the American Federation of Teachers and United Federation of Teachers sets school AI privacy and safety protections. AP reported that the standards would apply to schools with Microsoft contracts from 1 November. They restrict training on student and educator data, with a narrow safety exception, and prohibit selling data, advertising uses and features designed to create emotional dependency.

Why this matters: The useful development is the move from promises to enforceable protections. For UK schools, it offers questions to ask suppliers; it does not automatically rewrite our contracts.

📝 Teachers redesign assessment for the AI age

What it is: Reuters’ 28 October report, republished by AOL, draws on interviews with a dozen US teachers and education experts. It describes less take-home work, more classroom writing and discussion, and tasks that make pupils’ thinking visible.

Why this matters: A polished final answer tells us less when software can produce one. Seeing a pupil explain, draft and revise gives us more to work with. This is reporting on teachers’ responses, not evidence that one assessment model solves the problem.

🇭🇰 Hong Kong plans AI in the primary curriculum

What it is: Reporting on 30 October describes a provisional Information and Innovation Technology curriculum, with AI alongside coding and robotics. Phased implementation is planned from 2028/29, beginning with Primary One and Primary Four. This is a proposed framework, rather than a new subject already being taught everywhere.

Why this matters: It puts the curriculum question firmly on the table: what should children understand about AI as they grow up? My interest is in progression, including limits and judgement, alongside the practical skills. A timetable also needs trained teachers and useful resources behind it.

🔍 Claude’s invisible text watermark: a signal, not a verdict

What it is: A customer email published on 28 October says Anthropic will extend watermarking to four older Claude models on its API from 3 November. Anthropic explains that its watermark uses a statistical pattern in word choices, with no hidden characters added. Its text detector is in private preview for eligible organisations.

Why this matters: This is not a universal homework checker. Anthropic says short passages and light proofreading can provide too little signal; extensive rewriting can remove it. A result indicates likely Claude involvement, without establishing the pupil’s whole writing process. I’d still want the drafts and the conversation.

A little AI light relief before we go - It does not look like the role of teachers is disappearing any time soon!

Watch Husk’s video on X

’Till next week.

Mr A 🦾

Help a colleague save time by sharing this newsletter; distributing these ideas helps a friend get home on time and keeps our energy focused on what matters most: great teaching.

The tools and workflows mentioned are intended for professional productivity and educational enhancement. Users must ensure that any AI implementation remains compliant with their local data protection regulations and institutional safeguarding policies.