The number in the headline is eighteen. It is not a count of months, because nobody has measured the lag in months. It is the percentage of American public-school teachers who receive any formal guidance from their employer on how to use AI at work.
The comparison number is thirty-two: the percentage who use AI at work at least once a week.
Both come from Gallup, which is the only organisation running a repeated, properly sampled panel on this question anywhere. In March and April 2025 it asked 2,232 US public K-12 teachers about adoption. Sixty percent had used an AI tool for work that school year; 32% used one weekly. Weekly users estimated saving 5.9 hours a week, which across a 37.4-week school year is about six weeks of time. The most common uses were preparing to teach (37%), making worksheets and activities (33%) and adapting materials for individual students (28%). Between 57% and 74% said the work came out better, depending on the task, and fewer than 16% said it came out worse.
In February and March 2026 Gallup asked a fresh sample of 2,069 teachers a different question: does anyone tell you how to do this? Eighteen percent get formal guidance. Thirty-four percent get none at all, across the ten activities the survey measured. Another 48% get informal guidance, which in practice means a colleague's opinion. Fewer than 10% get formal guidance on any single specific activity. On the two uses with the highest stakes for a student, the gap is widest: 69% receive no guidance at all on using AI for one-on-one instruction or tutoring, and 58% receive none on grading and student feedback.
A third of a profession is doing something weekly that a fifth of it has any rules for.
The regulator has not been slow. It has been deferred
It would be easy to read this as regulators failing to keep up. In education specifically, that is not what happened.
The European Union classified AI in education as high-risk in the AI Act, explicitly, by name. Annex III point 3 covers AI used to decide admission to an institution, AI used to "evaluate learning outcomes, including when those outcomes are used to steer the learning process", AI used to assess what level of education a person should receive, and AI used to monitor students during tests.
Read the second one again, because it is the category. An adaptive learning engine that measures what a student got wrong and changes what it shows them next is a system that evaluates learning outcomes and uses them to steer the learning process. That is not an edge case of the definition. That is most of the product category.
The obligations attached to that classification are substantial: risk management, data governance, technical documentation, record-keeping, transparency, human oversight, accuracy and security. The Act entered into force on 1 August 2024 and those duties were originally due to bite on 2 August 2026. Under the Digital Omnibus political agreement they were pushed to 2 December 2027 for stand-alone Annex III systems, education included.
Nothing was softened. Sixteen months were added. Any company selling into Europe, or selling to a Gulf ministry that reads European rules as a template, has been handed runway rather than a reprieve.
UNESCO, meanwhile, published the first global guidance on generative AI in education in September 2023 — recommending an age limit for independent student use of these platforms, and noting that in most countries the absence of national rules leaves student data unprotected. That guidance is now approaching its third anniversary. It is guidance, not law, and it has no enforcement behind it anywhere.
Teaching about AI is not teaching with AI
The Gulf has moved faster than almost anyone on AI in schools, and it has moved on a different question than the one this article is about.
The UAE made AI a mandatory curriculum subject in government schools from reception through grade 12 starting in the 2025-26 academic year, across seven curriculum areas from foundational concepts to ethics and project design. Saudi Arabia introduced an AI curriculum across all levels of public education from the same year, built jointly by the National Curriculum Center, two ministries and SDAIA, with outcomes feeding into national student assessment. Stanford's AI Index records China and the UAE as the two countries that mandated AI education from 2025-26.
Those are real policies and they are ahead of most of the world. They are also policies about AI as a subject. They say nothing about AI as a method.
No Gulf ministry has published a learning-outcomes evaluation of an AI tutor. The distinction is easy to blur in a pitch deck and impossible to blur in a procurement meeting, and a founder who cites the UAE mandate as evidence that the region has validated AI-assisted teaching is citing the wrong document.
Nobody has measured the same two gaps here
The honest position on MENA is that we have Gallup's questions and none of Gallup's answers.
There is no survey of teacher AI adoption in Jordan, Egypt, Saudi Arabia or the UAE comparable to the American panel. The US figures of 60% and 32% should not be transplanted; they describe a different school system, a different device base and a different labour market for teachers.
What does exist for the region is a rapid scan published in January 2026 by EdTech Hub, commissioned by the UK's FCDO. Its findings on the governance side are consistent and unflattering. Ministries in Jordan, Iraq and Egypt have not embedded AI-focused training into national professional development. Adoption is concentrated among urban elites, private schools and well-resourced universities, while public-sector schools lag on infrastructure and ministry-led integration. Many countries in the region have no data-protection law under which student data collected by these platforms would be governed. Several widely used regional platforms depend on Western-developed AI engines, which is a licensing and continuity exposure rather than a pedagogical one.
The most telling thing about that brief is what it does not contain: a single learning-outcomes evaluation. A regional scan commissioned by a development agency, published in 2026, cites no effect sizes and no controlled trials of AI in education anywhere in MENA.
Meanwhile the students are not waiting. Four in five American high-school and college students use AI for schoolwork; only half of American middle and high schools have any AI policy, and just 6% of teachers say the policy they have is clear.
What a founder should do about it
Ship the governance artefact as part of the product. The buyer has no policy. The teacher has no guidance. If your product does not arrive with a usage policy template, a teacher-facing explanation of what the model does with student work, an audit log and a data-residency answer, someone in the ministry has to write all four before they can approve you. Whoever supplies them wins the evaluation.
Treat 2 December 2027 as a build deadline, not a legal one. Risk management files, technical documentation and human-oversight design take longer to retrofit than to build in. The sixteen-month deferral is the cheapest engineering time this category will ever get.
Do not sell a Gulf AI curriculum mandate as evidence for your tutor. It is evidence of political will and budget. It is not evidence that the method works, and the person across the table knows the difference.
Publish your own failure modes. In a market where nobody can independently verify anything, and where no regional evaluation exists to verify against, disclosing what your system gets wrong is the only trust signal that costs you something. That makes it the only one worth anything.
The gap in the headline is a US measurement. The version of it that matters to this region has never been taken.