Hey {{ first_name | human }},

Plenty to get through this week. Google is giving schools more control over AI, although there’s a rather important catch. The DfE wants your views on safer EdTech, and new maths tutoring research gives us something more useful than another impressive demo: results worth looking at carefully.

There are a couple of model releases to catch up on too. Let’s get into it.

🔎 Google’s AI switch has a catch. Gemini can be disabled separately, but keeping Search means keeping its built-in AI features.

🛡️ Safety certification is proposed. The DfE consultation closes on 4 February 2027. A safety badge would not generally tell us whether a product improves learning.

🧮 AI tutoring shows promise, with limits. A trial involving 1,179 pupils found stronger immediate performance than static support, but no clear gain on a broad standardised assessment.

🔎 Google gives schools an AI switch, but there’s a catch

What it is: On 3 November, Google announced changes to controls for Gemini, Gemini Notebook, Gemini in Classroom and Search. These services would be off for K–12 student accounts until administrators enable them. Google says schools will receive more detail before rollout; this is not a change already implemented everywhere.

Schools can keep Search while disabling Gemini. The catch is that Search’s own AI features are part of the package: opting out of AI altogether means leaving Search off for those managed accounts. Google also announced Guided Learning in Classroom and a Search pilot in selected US schools later this month.

Why this matters: “Can we turn AI off?” needs a more precise answer than a single switch. I’d want leaders and IT colleagues to check which services, accounts and age groups a setting covers, and what pupils would actually see. Choosing a search engine should leave room for a thoughtful conversation about how children learn to research.

🛡️ DfE proposes safety certification for EdTech and AI

What it is: Today, 10 November, the DfE launched a consultation for England on independent product safety certification. Proposed categories include pupil-facing generative AI and personalised learning tools, filtering products and technology processing children’s data at high risk.

The scheme would initially be technically non-mandatory. Proposed statutory guidance would make certification the usual route to assurance, while allowing justified alternatives supported by credible evidence. Making it legally compulsory would need new powers. Responses are due by 4 February 2027.

Why this matters: Independent checks could help schools ask better questions and reduce duplicated work. But the distinction matters: certification would assess safety, not generally educational effectiveness or overall quality, except where effectiveness directly affects safety. “Safe enough to use” and “worth using for this learning goal” are still separate judgements. I’d encourage schools and trusts to read the full proposals and share what would make assurance useful in practice.

🧮 Eedi and Google DeepMind put AI maths tutoring to the test

What it is: Findings published on 29 October cover a randomised trial with 1,179 pupils in 10 UK state schools, including 10 weeks of lessons. Pupils received static support, human tutoring or one of two AI tutoring versions. Every AI draft was reviewed by a human tutor before reaching a pupil.

After tutoring, pupils answered a similar question correctly around 56–58% of the time, compared with 43% after static content, with no clear difference between AI and human tutoring. But AI tutoring showed no clear advantage on the broader Star Maths standardised assessment. That analysis included the 916 pupils who completed both tests.

Why this matters: Doing better on the next question is encouraging. It does not establish lasting, transferable learning. This was also a structured system with human review, rather than an unsupervised general chatbot. I’d want longer-term evidence before making a bigger claim. Eedi’s full findings are refreshingly clear about those limits.

OpenAI’s Images 2.6, released on 4 November, brings faster generation, more consistent editing and GIF creation. Anthropic’s Claude Fable 5.2 and Claude Mythos 5.2 followed on 9 November, with claimed improvements in coding, reasoning and vision. Mythos access requires verification for relevant research uses. Useful developments to watch, but stronger technical performance is not evidence that these models will make a massive difference for teachers.

’Till next week.

Mr A 🦾

P.S. I've recently launched Enacture, a consultancy helping schools and trusts improve teaching, learning and professional development.

If your school is exploring how to use AI thoughtfully and effectively, whether that's developing an AI strategy, supporting staff or deciding which tools are worth using, I'd love to hear from you.

Get in touch through the Enacture website and tell me a little about what your school is working on. I'd be happy to arrange an initial conversation.

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.