← Back to Blog
🎓 AI in Education

AI in Education Digest — October 26, 2026

Dark navy graphic with the white serif headline Check the learning, not the score, underlined in teal. On the right, a tall solid teal column stands next to a short, empty, teal-outlined column topped by a large teal question mark. Footer: AI in Education Digest, 26 Oct 2026, derecktafuma.co.uk.

Seven stories, and most of them ask one question: is a higher score the same as more learning? A New York Times investigation found that fewer than half of the students at Alpha School, the AI-first private school network, hit its “learn twice as fast” promise in both maths and reading. A new research roundup reports that an AI tutor lifted homework scores but made no difference on an exam taken without it. In South Korea, about 600 students must resit a midterm after one boasted of scoring 99 with AI. Elsewhere, Oxford and the DfE start work today on how pupils’ data should (and shouldn’t) feed AI, India puts a 20-hour AI module into secondary schools, Bihar signs up 11 tech firms to train teachers, and Australia’s teachers’ union says AI has moved their workload rather than cut it. The takeaway for families: judge AI by what your child can do when the tool is closed.

1. NYT: fewer than half of Alpha School students hit its “2x” AI learning promise 23 Oct · 72h window

Source: Dana Goldstein, The New York Times · 23 October 2026

Alpha School is a for-profit network of private schools where academics take two to three hours a day on apps, with no traditional teachers. It markets “2x in two hours”. The Times reviewed Alpha’s own data for about 400 K–8 students from fall 2025 to early 2026. Most are high achievers (median at the 87th percentile nationally in maths and the 89th in reading), and 75% advanced at least as fast as national averages. But only 42% met the “2x” growth standard in both maths and reading. Of the students who started at or below the 50th percentile in maths, only 37% hit 2x. The software also reaches about 2,600 public-school students in Texas and South Carolina. Houston’s superintendent said pilot students spent longer than expected on the apps, and cognitive psychologist Daniel Willingham called the approach “naïve theories of psychology”. Alpha declined to answer the Times’ questions.

Why it matters: This is the most high-profile “AI replaces the classroom” model, and its own data shows the promise held for under half of pupils, and less often for those who started behind. For any AI learning app, ask: what does progress look like on an independent test?

X reaction: @Techmeme (13,222 impressions) · pushback from @RobDonnelly47: “The NYT article about Alpha school tried so hard to frame the results as negative even though they seem super impressive.” · @EdTechMuser on the “mic drop ending” by @DTWillingham.

Video angle: “An AI school promised your child would learn twice as fast. Its own numbers say fewer than half did.”

Navy bar chart titled Hit 2x? with two teal bars: both subjects 42 percent, low starters 37 percent. Alpha School K-8 private-school students meeting its 2x growth goal, fall 2025 to winter 2026; low starters are at or below the 50th percentile in maths. Source: Alpha data reviewed by The New York Times, 23 Oct 2026.

Share of Alpha School K–8 students meeting its “2x” goal. Source: Alpha data reviewed by The New York Times, 23 Oct 2026.

2. Research roundup: an AI tutor lifted homework, not the exam

Source: AIEd Research Weekly Update #1, AI in Education Knowledge Base · 24 October 2026 (research summary)

The roundup’s headline: “a score is not the outcome”. In a randomised trial, 153 proof-writing students at Carnegie Mellon got an AI proof tutor early or late. Homework scores rose 2.56 points while the tool was available, but midterm scores on an exam taken with no AI did not differ. Once everyone had access, every 100 extra chatbot messages went with a 3.02-point lower score on the next exam, and less confident students used it most. In a separate Munich trial (275 computer-science students), exercise scores were 71.84% with a hint-giving tutor, 57.50% with unrestricted ChatGPT and 29.85% with the open web, but gains in conceptual knowledge did not differ between the groups.

Why it matters: Story 1’s pattern, from controlled studies: AI can raise the mark on the work it helps with without raising what the student can do alone. In GCSEs and A-levels, the unaided test is the one that counts.

X reaction: No X reaction to these studies found in the window.

Video angle: “In a real university trial, an AI tutor pushed homework scores up, and exam scores stayed exactly where they were.”

Two navy cards side by side. The left card, outlined in teal, has a teal upward arrow and the words Homework up. The right card has a flat grey line and the words Exam unchanged. AI proof tutor trial with 153 Carnegie Mellon students: homework up 2.56 points, no difference on the unaided midterm. Source: Chen et al., summarised in AIEd Research Weekly Update number 1, 24 Oct 2026.

Homework went up; the unaided exam didn’t move. Source: Chen et al., via AIEd Research Weekly Update #1, 24 Oct 2026.

3. Oxford and the DfE launch a project on how pupils’ data should be used for AI

Source: Kristina Khoo-Rhodes, University of Oxford Department of Education · 26 October 2026

A six-month project, Publicly Funded Pupil Data as a Strategic Asset and Public Good, launched today. It was co-designed with the Department for Education and is funded by an ESRC React Award. It will look at how anonymised pupil-generated data could be used for AI research and tools, “while first and foremost, protecting children’s rights and maintaining public trust”. Partners include the DfE, Ofqual and Oak National Academy, with support from Baroness Beeban Kidron. Promised outputs include a governance framework for pupil data, guidance on a “Pupil Data Commons”, and recommendations for future AI assurance and certification.

Why it matters: Every AI tool in school runs on data, often your child’s. This is the first sign of England setting rules on whether pupil work can train AI.

X reaction: None yet (published this morning).

Video angle: “Your child’s schoolwork is data. Who gets to train AI on it? England just started deciding.”

4. India: NCERT releases a 20-hour AI module for Classes 9–12

Source: PTI via The Hindu · 25 October 2026 · also Education Times · 25 October 2026

India’s national curriculum body has released an activity-based, 20-hour introductory module on computational thinking and AI for Classes 9 to 12. It was developed with IIT Madras and IIT Jodhpur and needs no prior knowledge. It covers how machines learn from examples, detecting deepfakes, data privacy, misinformation, fairness and checking AI output, and can be taught with or without devices. NCERT wants students to be “careful, confident, creative, and responsible users and evaluators of AI systems, rather than relying on its outputs without verification”.

Why it matters: A national system is teaching teenagers to question AI, not just use it. A useful comparison for England’s GCSE years.

X reaction: @EconomicTimes (5,054 impressions) · @India_tech_news: “Students will learn how AI works, spot deepfakes, protect data privacy and use AI responsibly.”

Video angle: “Twenty hours, no devices needed, and the main lesson is: don’t trust the AI until you’ve checked it.”

5. Australia: teachers’ union tells the Senate that AI has “shifted”, not cut, workload

Source: Adam Vidler, Nine.com.au · 24 October 2026

Nicole Calnan, federal secretary of the Australian Education Union, told the Joint Senate Committee on Artificial Intelligence that “Governments have promoted AI as a way to reduce teacher workload yet our members experience and their available research is that it shifts their time into checking and reworking outputs.” She warned that leaving AI training to the companies selling AI “risks access following existing patterns of disadvantage”, and said reports to the eSafety Commission of sexualised deepfakes targeting students and teachers had doubled in 18 months. A nine.com.au poll of 1,209 people found 28% want chatbots kept out of schools entirely and 85% don’t want them in primary schools.

Why it matters: UK ministers make the same “AI will save teachers time” pitch. Australia’s biggest teachers’ union is reporting time spent checking AI instead.

X reaction: No relevant X reaction found.

Video angle: “Teachers were promised AI would cut their workload. Their union says it’s just turned them into AI checkers.”

6. South Korea: about 600 students resit a midterm after an AI-cheating boast 23 Oct · 72h window

Source: Choi Youngchan, The Asia Business Daily · 23 October 2026 · also Seoul Economic Daily · 23 October 2026

A Konkuk University professor voided an online midterm, held on 19 October for a general-education course of about 600 students, after a student posted on a campus forum that they had scored 99 using AI answers. The professor wrote that “the fairness and ability of the exam to distinguish between students were not adequately ensured because some students cheated, including by using AI”. Everyone resits in person on 14 December.

Why it matters: One shortcut cost 600 people their result. Unsupervised online tests are on the way out, and students who actually learned the material lose nothing when that happens.

X reaction: @yonhaptweet (5,499 impressions) · @KBSnews (3,086) · @zzeong__: “와 599명은 무슨죄냐, 불쌍해…..” (roughly: “What did the other 599 do wrong? Poor them.”)

Video angle: “One student bragged about scoring 99 with AI. Now 600 people have to resit.”

7. Bihar signs up 11 IT firms, including Google Education, to train teachers in AI

Source: B K Mishra, The Times of India · 25 October 2026

Bihar’s education department has partnered with 11 IT companies, including Google Education, to train government school teachers in IT and AI in phases through the state’s SCERT, with arrangements due by January. Computer education is set to become compulsory from Class VI in government schools next academic session, starting with smart classrooms in 700 model schools, and university faculty face mandatory AI training.

Why it matters: Mass AI teacher training delivered by tech firms is exactly what Australia’s union (Story 5) warns about.

X reaction: No X reaction found in the window.

Video angle: “Eleven tech companies are about to train a state’s teachers in AI. Should the people selling the tools be the ones teaching them?”

🎬 Best story for today's video

Check the learning, not the score (Stories 1 and 2)

The Alpha School data and the Carnegie Mellon trial tell the same story: AI can lift the visible number without lifting what the child can do alone. One habit answers both: after any AI-helped task, do one question with the tool closed. Runner-up: the Konkuk resit.

“An AI school promised double-speed learning, and a university trial saw homework scores jump. In both cases, the test with the AI switched off told a different story.”

AI in educationAI tutorsAlpha Schoolhomeworkexamspupil dataAI literacyteacher workloadacademic integrityparents

Want this digest in your inbox? A short roundup of what’s actually changing in AI and learning.

Join the newsletter →

Test it with the tool closed

Get The Revision Lab’s free GCSE History revision sheet: key people and dates for Black History Month on two A4 pages, with a timeline, exam tips and a quick quiz with answers, so your child can check what they know without AI. The GCSE Maths Mock Booster is live.

Get the free revision sheet The Revision Lab