Term is days away and the AI-in-education story just split in two. On one side, institutions are pouring money and free tools into AI faster than ever. On the other, some of the most credible names in science and child safety spent this week drawing hard lines around it — including NASA, which has just told students they cannot use AI at all.
1. NASA bans generative AI from its 2027 Student Launch challenge
Source: EdTech Innovation Hub · 26 August 2026
NASA has opened applications for the 2027 Student Launch challenge with a rule that stands out sharply against the rest of this year’s education news: students may not use generative AI or AI tools for any challenge-related work. The ban explicitly covers browser-based generative AI, AI-assisted coding tools and AI-generated imagery, and forbids uploading challenge content to AI systems for processing.
It applies across both divisions — college and university teams, and middle and high school teams — and sits alongside NASA’s long-standing rule that students complete 100% of the project themselves, with adult mentors restricted to safety-critical tasks such as motors and ejection charges. Applications close on 14 September, teams are announced on 1 October, and Launch Day is 10 April 2027 in Huntsville, Alabama. University teams face an autonomous post-landing imaging payload that must locate and photograph a passive visual target without GPS, within 15 minutes of landing.
Expect “AI-free zones” to become a normal part of assessment design, not a protest against technology.
2. Common Sense Media puts ChatGPT for Teens under independent test
Source: EdTech Innovation Hub · 26 August 2026
Common Sense Media’s Youth AI Safety Institute is preparing an independent risk assessment of OpenAI’s new teen experience, which activates automatically when ChatGPT knows or suspects a user is under 18. Researchers will focus on three things: whether the Responsible Homework Reminder actually stops students shortcutting an assignment, whether expanded parental alerts fire reliably, and what the new customisable voice and personality options do to teenagers emotionally.
The organisation previously rated ChatGPT “High Risk” overall and “Unacceptable Risk” for mental health support. Developmental behavioural paediatrician Jenny Radesky singled out the voice modes — Professional, Friendly, Candid, Quirky and Cynical — as her “biggest concern”, questioning whether an authoritative tone might reduce a teenager’s trust in their own answers. Robbie Torney and Radesky wrote that the features are “well intentioned, but we need additional transparency and external testing to confirm that they perform as intended.”
If you are a parent, the useful question is not “is it safe?” but “is the parent account actually linked?”
3. A 6,997-student trial finds AI tutoring helps after mistakes — but costs time
Source: EdTech Innovation Hub, reporting an NBER working paper · 24 August 2026
A randomised trial across 20 schools in Hamilton County, Tennessee, put a guard-railed AI maths tutor in front of 6,997 middle school students during a normal 50-minute lesson. The clearest effect appeared immediately after an error: among students in a mastery workflow, being assigned the AI tutor raised the chance the next attempt was correct by 8.5 percentage points and cut the number of extra attempts needed by 0.96. It also added 2.88 minutes to reach that correct answer.
A week later the learning gain was real but modest. On the practised question, 40.2% of AI students answered correctly against 37.0% of students on the platform without AI — a 3.2 point difference the authors call suggestive rather than definitive, with a p-value of 0.065. Co-author Alp Süngü summarised it on LinkedIn: “AI improved post-mistake accuracy! But it also slowed students down.” Crucially, the tutor was purpose-built not to hand over answers, with responses screened for correctness before delivery. This is not a study of teenagers using a general chatbot.
The most quoted line will be the 8.5 points. The most important one is that it is a working paper, not peer reviewed.
4. 70% of US teens use AI for schoolwork — and almost nobody has taught them how
Source: EdTech Innovation Hub, reporting Common Sense Media · 21 August 2026
Common Sense Media’s report Teens in the AI Era, based on a NORC survey of 1,017 teenagers aged 13 to 17 across all 50 states, found 70% now use AI for schoolwork. Among those users, 77% use it for genuine support such as brainstorming, checking work and getting feedback — while 63% also get answers in some form, split between fixing the answer (35%), rewriting it to sound like themselves (31%) and using it as-is (25%).
The teaching gap is the sharper finding. Only 27% say a teacher has explained what AI is and how it works, 30% have covered AI safety, and just 26% have been taught how to judge whether an AI answer is accurate. Thirty-five per cent do not understand or are unsure of their school’s AI rules. Researcher Michael Robb noted that “teens are more deliberate than the debate assumes… it’s often the same kids doing both.” Asked what skills matter, students ranked working with others (55%) and fact-checking (50%) far above knowing how to use AI (21%).
Also telling: 44% have had a tool blocked at school, and 59% of those simply switched to a personal device.
5. US Education Department tells schools to judge EdTech by outcomes, not screen time
Source: EdTech Innovation Hub · 25 August 2026
A Dear Colleague Letter issued on 20 August asks US states and districts to stop evaluating classroom technology by how much screen time it involves and start evaluating it by measured learning impact, favouring products with demonstrated positive outcomes including randomised controlled trials. Assistant Secretary Kirsten Baesler drew the distinction plainly: “Cell phones and algorithm-driven social media distract students from learning… In contrast, instructional technologies should be judged by their impact on learning, not simply whether they happen on a screen.”
The letter tells vendors to publish independent evaluations, give classroom-grounded implementation guidance and be transparent about limitations — and tells schools to change course when a tool is not working and remove it when the shortcomings persist. Louisiana, Arkansas, Indiana, Michigan and Texas are cited as exploring outcome-based contracting. One line does the heavy lifting: “Responsible design is the floor.”
Outcome-based procurement is unglamorous. It is also the fastest way to kill the AI products that do not work.
6. AWS commits more than $500m to university cloud and AI training
Source: EdTech Innovation Hub · 25 August 2026
AWS has committed more than $500 million in resources to a Student Rewards programme open to verified higher education students worldwide aged 18 and over, delivered through its Builder Center platform. The package is worth up to $579 per student: twelve months of AWS Skill Builder premium (valued at $449, covering 900-plus courses, labs and certification prep), $30 in AWS credits and a $100 foundational certification exam voucher.
The rewards are earned rather than granted. Seven badges unlock $10 in credits, fourteen unlock another $20, and twenty-one unlock the exam voucher — with badges awarded for publishing articles, commenting and staying engaged in the community. Kim Majerus, AWS VP of Global Education, wrote: “I got into tech at 16 because someone gave me a chance to explore something I never thought I’d love.” Verification runs through SheerID, and the programme is global, so UK undergraduates qualify.
Read it as a recruitment funnel with real benefits attached, not a donation.
7. Anthropic opens Claude Academy with free AI courses — no sign-in required
Source: EdTech Innovation Hub · 24 August 2026
Claude Academy went live on 20 August with free courses, tutorials and use cases, and no login needed to access them. The flagship path is AI Fluency: Framework & Foundations — roughly four hours across 14 lessons and a quiz, built on a four-part framework of Delegation, Description, Discernment and Diligence. A separate AI Capabilities and Limitations course runs 13 lessons on knowledge, working memory, steerability and context limits.
The framing is unusually education-shaped for a vendor. Anthropic states that “AI instruction should increase agency and empower learners to expand their capabilities”, and teaches learners to “verify in proportion to the stakes” and to disclose AI contribution to colleagues and customers. Account holders can track progress and earn completion badges for LinkedIn.
Vendor-produced and product-linked — worth using, worth pairing with something independent.
NASA just banned AI from its student rocket challenge
Every other story this week is about adding AI — free courses, $500m of credits, new teen products. NASA went the other way, and that contrast is the whole video. It is surprising, it is easy to explain in ten seconds, and it lands directly on the question every student and parent is actually asking in the last week of the holidays: when am I allowed to use this, and when am I not?
NASA just told thousands of students they are not allowed to use AI. No ChatGPT, no AI coding tools, no AI images. In a year when every school on earth is racing to add AI, the space agency did the opposite — and the reason is smarter than it sounds.
The thread running through all seven
Read together, this week is not a fight between pro-AI and anti-AI. It is a fight about where AI belongs. NASA and the US Education Department are both, in their own way, asking the same question: what is this activity actually meant to measure, and does the tool help or hide it? The Tennessee trial answers it with numbers — a tutor built to withhold answers produced better recovery from mistakes, and a tutor built to hand them over would not have.
The gap, as the Common Sense data shows, is that almost nobody has taught students to make that judgement themselves. Free material from Anthropic and AWS helps with the how. It does not help with the when. That part is still on schools, and September is a fortnight away.
Get this digest every morning. A five-minute roundup of what is actually changing in AI and learning — straight to your inbox.
Join the newsletter →