Today's headline is proof that most "AI cheating" isn't clever — it's copy-paste laziness, and one professor's viral trick exposed it perfectly. Around that story: a national cheating scandal in Mexico that's reached the president's desk, Europe's AI Act rules technically arriving for schools (before most of them got delayed), a new Florida district AI policy, fresh Stanford data on the adoption-versus-policy gap, and a genuinely hopeful result from an AI tutoring pilot in Sierra Leone.
1. Professor's hidden "Madagascar" trick catches 32 students cheating with AI
Source: The Register / Fox News / Futurism · August 1–3, 2026
Alcorn State University history professor Jason Gibson buried invisible white-text instructions inside his midterm essay prompt, telling any AI tool reading it to insert the word "Madagascar" somewhere in a way that made no sense. Students who pasted the prompt into a chatbot and submitted the answer without reading it got caught immediately.
Across his two classes, 32 of Gibson's 35 students failed a portion of the exam after their polished, AI-written answers contained bizarre, nonsensical references to Madagascar. Gibson's TikTok series walking through the trick went viral this week, sparking a wider conversation about how easy — and how cheap — it is to catch careless AI use without any detection software at all.
Expect more teachers to start quietly testing their own hidden-instruction traps this term.
2. Mexico's top university annuls thousands of exams in a massive AI cheating scandal
Source: CNN / Mexico News Daily · August 2, 2026
UNAM, Mexico's most prestigious university, ran its entrance exam online for the first time this year and was flooded with suspiciously high scores. Around 3,000 of the roughly 158,700 tests taken have been formally annulled so far, and one AI specialist estimates as many as 75,000 test-takers — nearly half — may have used AI to cheat.
UNAM has suspended new enrolment while it investigates and plans an in-person "control exam" for around 58,000 applicants who passed this year or in the past five years. The fallout has become a national political story: President Claudia Sheinbaum, herself a UNAM physics graduate, has publicly suggested involving the attorney general.
Worth watching whether other Latin American universities quietly reconsider online entrance exams this year.
3. The EU AI Act's high-risk deadline for schools arrives — then gets deferred anyway
Source: Travers Smith / DLA Piper / EU legal trackers · August 2, 2026
The EU AI Act's high-risk compliance deadline for education — covering AI used in admissions, grading, and academic-progress monitoring — technically arrived on August 2. But a last-minute "Digital Omnibus" amendment pushed the heaviest documentation, risk-assessment, and human-oversight obligations out to December 2027.
A narrower rule is already in force: institutions must disclose to students when they're interacting with AI. The Act's extraterritorial reach means even non-EU institutions serving EU students or running remote programmes accessible to EU users remain potentially in scope.
Worth bookmarking December 2027 as the date this story resurfaces for real.
4. Florida school district passes its first-ever classroom AI policy days before term starts
Source: Spectrum News 13 (Orlando) · July 29, 2026
Orange County Public Schools approved its first formal AI policy after surveying roughly 20,000 parents and teachers and holding student town halls across the district. Teachers must publish written AI guidelines at the start of the year, and students can use AI for background research but not for graded work without written permission.
Notably, the district is banning the use of AI-detection tools on student work entirely, arguing they're unreliable. One board member voted against the policy specifically because it lacks a parental opt-out — a fight likely to resurface in other districts.
Orange County's new school year — and this policy — begins August 11.
5. Stanford's 2026 AI Index: 4 in 5 students use AI, but barely any schools have a real policy
Source: Stanford HAI, AI Index 2026
Stanford's flagship AI Index reports that roughly 80% of U.S. high school and college students now use generative AI for schoolwork. Yet only about half of middle and high schools have any AI policy at all, and just 6% of teachers say the policy they do have is actually clear.
The report frames this as education entering the AI era "not through orderly reform but through mass student adoption, patchy policy, and uneven teacher preparation" — a gap that's widening pressure on schools to catch up fast.
A good one to revisit once more districts publish their back-to-school AI policies this month.
6. Google DeepMind's AI tutor delivered a year of math progress in just 8 weeks
Source: Forbes (interview with DeepMind researcher) · July 2, 2026
In a classroom pilot in Sierra Leone, Google DeepMind's "Guided Learning" AI tutor helped students gain more than a full year's worth of math schooling in only eight weeks, according to the researchers who ran the trial.
It's one of the more concrete pieces of evidence that AI tutoring can produce dramatic learning gains in real, under-resourced classrooms rather than just controlled lab studies — with obvious implications for school systems short on qualified teachers.
Worth a follow-up once DeepMind publishes fuller trial data beyond this initial report.
This Professor Used ONE Hidden Word to Catch 32 AI Cheaters
Of everything in today's digest, this is the funniest, most rewatchable, and most immediately useful story — a genuinely clever trick any teacher can copy, with a built-in twist ending that makes for a perfect hook and thumbnail.
"A history professor hid one random word in his exam instructions — and it just exposed almost every student in his class for cheating with AI. Here's exactly how he did it, and why it works on basically anyone who doesn't read their own homework."
Want tomorrow's digest before anyone else?
Get the daily AI-in-education briefing straight to your inbox — no fluff, just the stories that matter.
Subscribe to the newsletter →