The AI labs want your child's teacher, not your child

Most parental worry about AI still points at a child’s own screen. Which chatbot is open, what it said back, whether the essay is really theirs. Reasonable questions, all of them.

But the biggest education announcements from the AI labs this month were not aimed at children at all. They were aimed at the adult standing at the front of the room.

The pitch this month went to teachers

On 14 July, Anthropic launched Claude for Teachers: free premium access for verified K-12 educators in the United States, free for a year to anyone who enrols before 30 June 2027. The pitch is operational rather than magical. Lesson planning mapped to academic standards in all fifty states, differentiating the same lesson for different readiness levels, reading class data, plus connections into evidence-based curricula such as OpenSciEd and Illustrative Mathematics and nine existing K-12 classroom tools. Anthropic states that “Claude for Teachers data is not used for model training purposes” and that student data is covered by a K-12 data processing addendum written to comply with the American federal student-records law. Chalkbeat reported the same day that a district pilot begins in Detroit Public Schools Community District next school year.

Google got there a few weeks earlier. At the ISTE conference on 25 June it announced study notebooks, where a student uploads their own material, takes a diagnostic quiz, and gets short lessons that reshape themselves as the quiz results come in. Free full-length ACT and GRE practice tests are coming through a Princeton Review partnership. The teacher-shaped part is the important one: teachers can assign a study notebook grounded in their own class materials inside Google Classroom, then see where the class or a particular child is stuck.

Two details in that post deserve a parent’s attention. Google describes its Guided Learning feature as “proven to drive positive learning outcomes” without pointing to the study, which is a strong claim to make without a citation. And the rollout line reads: “Study notebooks are rolling out now for users 18 years and older in their personal accounts, and will be available for all school-issued accounts, including for users under 18, in the coming weeks.” Anthropic’s free offer stops at the American border and at adults. Google’s does neither.

Why the teacher, and why now

Because that is where the growth is. Chalkbeat, citing Education Week survey data, reported that 61% of teachers said they used AI in their work in 2025, up from 32% in 2024. Teachers went from a minority to a majority in a single year.

The labs have been building toward this for a while. In July 2025 the American Federation of Teachers announced a National Academy for AI Instruction with Microsoft as lead partner and OpenAI and Anthropic alongside, a $23 million programme running out of a Manhattan building, aiming to train roughly 400,000 educators over five years and reach more than 7.2 million students. Microsoft’s Brad Smith framed it as a two-way street: “This partnership will not only help teachers learn how to better use AI, it will give them the opportunity to tell tech companies how we can create AI that better serves kids.” The union’s president, Randi Weingarten, put the guardrail in the same release: “The direct connection between a teacher and their kids can never be replaced by new technologies.”

Our own read, offered as opinion rather than reporting: a teacher who plans every week inside a tool is a far stickier customer than a teenager who tries a chatbot twice and drifts off. Winning the teacher wins the child by default, and it does so without ever asking the child’s parent anything.

Singapore went the other way

A Singapore parent reading all this should know that the route into their child’s classroom is narrower here, because MOE built its own tools instead of buying a lab’s.

On the teacher side, the Student Learning Space carries an Authoring Copilot that drafts lessons and assessment items from a teacher’s rough idea, and a Data Assistant that reads a whole class’s responses and surfaces the common misconceptions. On the student side, the Learning Assistant “helps students learn by asking guiding questions” rather than handing over answers, is available from Primary 4, must be supervised through Primary 6, has to be switched on by the teacher, and caps how many turns a child gets. The Adaptive Learning System that personalises maths practice is described plainly by MOE as “based on machine learning and not a generative AI”. The whole thing sits under four stated principles: Agency, Inclusivity, Fairness and Safety.

The age policy is deliberate. In a forum letter reply on 13 April 2026, Lee Lin Yee, Divisional Director of Educational Technology at MOE, wrote that “technology will be introduced in age- and developmental-appropriate ways” and acknowledged that “there are risks when young children use AI”. Lower primary stays print-first. Structured AI use starts at Primary 4, in class, under supervision.

The people building it talk the same way. In a GovTech piece last December, SLS product lead Harish said “we don’t shy away from using AI, to be honest. But we use it in a very deliberate and purposeful way”, and Lee Lin Yee offered the line worth taping to a fridge: “With more tech, we need more human connections.”

The honest limits

None of this makes the school AI good by default. MOE’s own guidance tells teachers to verify what the Authoring Copilot produces, because it can generate content that is inaccurate, incomplete or off-target. Daniel Buck of the American Enterprise Institute, quoted by Chalkbeat, warned that if teachers hand over the thinking parts of their job, “don’t be surprised when classroom community and academic outcomes rapidly deteriorate”. And a free tool that a whole profession learns on is a commercial position as much as a public good, whoever is offering it.

Disclosure: Mentus AI builds an AI mentoring product for children, so we have a commercial interest in this market.

So the question for the next parent-teacher meeting is not “does the school allow AI”. Ask which tool your child actually touches, who decided to switch it on, and what it does when your child gets something wrong. That last one matters most. A tool that supplies the answer and a tool that asks one more question produce very different children over a school year, and the research on that gap is not subtle.