Term starts on Monday for most of England, and the AI story in education has quietly moved past “should we allow it?” Adoption is now near-universal, the first serious evidence on AI tutoring is landing, and from 1 September AI stops being an IT question and becomes a statutory safeguarding one. Here are the seven stories worth knowing this morning.
1. 86% of students and 85% of teachers used AI last year — and researchers have a name for the downside
Source: News4JAX (WJXT) · 28 August 2026
A recent report cited by News4JAX found that 86 percent of students and 85 percent of teachers used AI during the 2024–25 school year. That is not a pilot programme. That is the default. The reporting frames AI as genuinely useful for brainstorming, studying and feedback — while flagging that it also makes it very easy to skip the part where the learning actually happens.
The specific worry has a name. A study in the British Journal of Educational Technology, referenced in the piece, found that tools like ChatGPT may promote what researchers call “metacognitive laziness” — a dependence on the technology that gets in the way of deeper learning. The practical advice for parents in the report is refreshingly concrete: be specific about what is off-limits, insist your child can explain any answer in their own words, and make “verify it” part of the routine. One tell to watch for this term — if a child’s writing suddenly sounds nothing like them, overly formal or stuffed with words they would never use, AI probably did more of the work than it should have.
The rule that beats every ban: AI can help you learn, but it cannot do the learning for you.
2. New research: AI tutoring works — but mostly when a human stays in the loop
Source: Education Week / Stanford National Student Support Accelerator · 28 August 2026
Two randomised controlled trials summarised by Stanford’s National Student Support Accelerator found that AI embedded in live, chat-based maths tutoring can improve student outcomes. In the Google/Eedi Labs study of LearnLM, supervising tutors approved 76.4 percent of the AI’s responses with little or no editing, and students who worked with LearnLM went on to succeed on harder follow-up topics 66 percent of the time, versus 61 percent for human tutors alone and 56 percent for static hints.
The second study, on Stanford’s Tutor CoPilot, randomised 1,000 elementary students to a human tutor with or without AI suggestions. Students in the CoPilot condition were four percentage points more likely to reach topic mastery — rising to nine points for those assigned the least experienced tutors. The mechanism matters: tutors using CoPilot were ten percentage points more likely to prompt students to explain their thinking, rather than fall back on generic encouragement. Education Week’s round-up on the same day lands on the caveat: AI cannot manufacture attendance, sustained contact over time, or a relationship.
The gains were biggest for the weakest tutors. That is a levelling-up story, not a replacement story.
3. From Monday, AI is part of the statutory safeguarding system in English schools
Source: Department for Education — Keeping Children Safe in Education 2026 · in force 1 September 2026
KCSIE 2026, the statutory safeguarding guidance every school and college in England must follow, comes into force on 1 September. This year’s edition pulls artificial intelligence firmly into the online-safety chapter and connects it directly to filtering and monitoring duties, referencing the DfE’s Generative AI: product safety expectations to explain how those requirements apply to generative tools in an educational setting.
The practical effect is that AI-enabled harms — inappropriate AI-generated content, unsupervised pupil interactions with AI systems, and data-protection exposure — are now mainstream safeguarding business rather than an IT footnote. Ofsted inspectors may reasonably ask how a school is managing those risks. That lands awkwardly against the adoption picture: survey work this year suggests roughly half of schools still have no AI policy for staff or students, and around two-thirds have none specific to pupil use.
If your school’s AI policy is still a draft in someone’s inbox, Monday is the deadline.
4. Universities are putting an AI-literacy gate in front of AI access
Source: Columbia aiX Weekly / UCCS · 19 August 2026
The University of Colorado Colorado Springs opened ChatGPT Edu to all enrolled students on 14 August — but accounts are only provisioned after students complete an “AI Literacy Foundations” course in Canvas. Staff and faculty who already had access must complete the same course by 25 September or have their access suspended. The institutional pitch is higher usage limits, university privacy controls, and a commitment that conversations and files are not used to train OpenAI’s models.
Columbia’s aiX Weekly, which flagged the move, makes the sharper point: universities have become much better at providing AI than at measuring what students learn with it. Seats provisioned and prompts submitted are easy dashboard metrics. Whether a student can explain their reasoning, spot a bad answer, or perform when the tool is taken away requires a completely different kind of evaluation — one that has to be designed before term begins, not after.
Course completion measures implementation. It does not measure judgement.
5. New South Wales asks whether unsupervised take-home assessment still works
Source: NSW Government · 10 August 2026
The New South Wales government announced a review of AI’s effects on student learning and the integrity of the Higher School Certificate, asking its assessment authority to consider pausing unsupervised take-home tasks. It is a school-sector decision, but the underlying question travels straight into universities and into every GCSE and A-Level coursework component.
That question is simple and uncomfortable: if assistance cannot be observed, what does a finished piece of work actually tell you about the learner? A large study published in Science in May, covering more than 95,000 students at 20 US public research universities, argued exactly this — that widespread AI use weakens the validity of assessment when submitted work is treated as direct evidence of individual learning.
Watch this one. Where NSW goes on the HSC, other exam boards tend to follow.
6. The faculty confidence gap: 74% suspect AI essays, only 21% know what to do about it
Source: College Board research brief, 2026 · fielded summer 2025
More than 3,000 US faculty responded to a College Board survey on generative AI in higher education. Seventy-four percent believed students were using AI to write essays or papers, and 92 percent reported ethical concerns about plagiarism or academic dishonesty. Only 21 percent said they felt confident and had a clear strategy for guiding AI use in their classroom.
That gap — near-universal suspicion, minority confidence — is the real story of this academic year. It also mirrors the UK picture, where survey work suggests around three-quarters of teachers have had no meaningful AI training or guidance from their school. Policy documents are multiplying much faster than practical capability.
Suspicion is at 74%. Competence is at 21%. That gap is where this year’s chaos lives.
7. 134 AI-in-education bills, 31 states, one very uneven map
Source: MultiState legislative tracking · 2026 session
Across the 2026 US legislative session, 134 bills relating to AI in education have been introduced in 31 states, covering data privacy, classroom-use restrictions, curriculum integration, graduation requirements and teacher training. The approaches diverge sharply. Oklahoma permits AI in schools only under educator supervision with human review, bans its use for high-stakes decisions, and requires annual parent disclosure. New York restricts classroom AI use to ninth grade and above, with carve-outs for diagnostics and special education interventions.
Meanwhile the American Federation of Teachers has called for bans on all screen use below grade 3 and on all student-facing AI, including digital tutors, in elementary schools. State boards are expected to move this year from issuing guidance to monitoring whether any of it is actually being implemented.
Guidance was 2025. Enforcement is 2026.
86% of students now use AI — and scientists say it is making them worse at thinking
Story one wins on every count: it has a hard number people will repeat, a phrase nobody has heard before (“metacognitive laziness”), a genuine fear parents already have, and — crucially — a fix you can deliver in ninety seconds. It also lands on the exact weekend before term starts, when every parent watching is already thinking about the year ahead.
86% of students and 85% of teachers used AI last school year. But researchers have a name for what it is doing to your brain, and it is not flattering: metacognitive laziness. Here is the one rule that fixes it.
Want this in your inbox every morning?
One short email a day: what happened in AI and education, and what to actually do about it.