Only 6% of teachers say their school’s AI policy is clear

Four in five American students already use AI for schoolwork, and half of their schools have no policy at all. Stanford’s data shows that the fastest adoption is happening well away from the countries writing the rules.

Four out of five American high school and college students now use AI for schoolwork. Only half of middle and high schools have any AI policy, and of the teachers working inside the ones that do, 6 percent describe the policy as clear.

The education chapter of Stanford’s AI Index sets that gap out, and it runs wider than almost anything else the report measures. We have covered two other chapters of the same 423-page document, on the gap between experts and the public, and on how China has nearly closed the gap on capability.

The figure that makes it concrete is not about cheating. Forty-seven percent of students said they wanted to use AI for a task but were not sure whether they were allowed to.

Roughly half the student population is guessing at the rules, which is a governance failure rather than an integrity one, and it falls on the institution rather than the child.

What students actually do with it is mundane. American high schoolers reach for AI to find sources (51 percent), edit or revise essays (50 percent), and brainstorm (50 percent).

University students worldwide report using it to understand concepts (56 percent), research (52 percent), generate ideas (46 percent) and edit essays (38 percent).

None of that is the plagiarism scenario schools spent three years writing policies about, and most of it is invisible to the tools built to detect it. Essay editing is the one worth watching, because MIT researchers found assisted writers showing lower brain engagement and performing worse once the tool was taken away.

The international picture inverts the usual story about who is ahead. Among university students surveyed in 2025, generative AI use runs at 95 percent in Indonesia, 89 percent in Australia and Turkey, 87 percent in Malaysia, 84 percent in Canada, 83 percent in Kenya, 79 percent in Brazil and 77 percent in Spain.

The United Kingdom and the United States both sit at 67 percent, at the bottom of the same table. Year-on-year increases since 2023 have ranged between 20 and 61 percentage points depending on the country.

Europe’s response has been to experiment at national scale rather than to legislate. Estonia’s AI Leap programme is running with 20,000 students and 3,000 teachers through 2025-26.

Greece has partnered with OpenAI to train secondary teachers on classroom use. Malta went further still and gave every citizen a subscription, conditional on finishing an AI literacy course written by the University of Malta.

South Korea supplies the counter-example as it launched AI textbooks in primary schools in March 2025 and reversed the policy after opposition from parents and teachers. Nothing in the report suggests the reversal was about the technology working badly.

China and the United Arab Emirates have gone the mandatory route. China requires it with minimum instructional hours and grade-level progression in Beijing, Guangdong and Hangzhou from the 2025-26 school year, and the Emirates has mandated it across all grades from the same date.

Against that, about 93 percent of countries teach computer science in some form, but only 30 percent require it, and 63 percent leave it as an elective available in some schools rather than all.

The weak point everywhere is the teacher. State-level guidance in the US recognises that AI teacher training matters and then provides no standards for the programmes and no funding for them, which leaves the work to whoever turns up offering it.

That vacuum has an occupant: Google, Microsoft and OpenAI have built their position in American classrooms partly through teacher training schemes and industry-written courses.

Some of what arrives is not opt-in at all, as with the AI summaries in Google Search that a child safety group judged an unacceptable risk to students and that schools cannot switch off.

Where districts have pushed back, it has sometimes been on grounds nobody drafted a policy for, such as the New York school that paused a classroom robot over its manufacturer’s corporate parentage.

On skills, the report measures how often AI competencies appear in job profiles relative to the global average. India leads at 3.0, and the United States follows at 2.0. Germany is the strongest European showing at 1.83, ahead of the United Kingdom at 1.55, France at 1.53, Spain at 1.47, Italy at 1.22, and the Netherlands and Poland level at 1.14.

Every European country in the set sits above the global average and well below India.

Split those figures by gender, and they look worse. German men register 1.93 against 1.06 for German women. British men are at 1.55 and British women at 0.94, which puts women in the UK below the global average in a country that is comfortably above it. Canada shows 1.63 against 0.95.

The same pattern holds in degree data, where women take 36 percent of AI software master’s degrees and 31 percent of bachelor’s, though Costa Rica and Latvia reached parity at doctoral level.

The supply side is moving in two directions at once. Computer science enrolment at American four-year universities fell 11 percent between 2024 and 2025, while master’s graduates in AI software fields rose 17 percent.

New AI PhDs across the US and Canada increased 22 percent from 2022 to 2024, and all of that growth went into academia rather than industry, reversing a decade in which the flow ran the other way.

The report is candid about what it cannot see. Comparable OECD data runs only to 2023 and excludes India, China and much of Africa. No standardised accounting of AI course offerings exists at school level anywhere, so computer science enrolment stands in as a proxy for something it does not measure.

Stanford’s own summary is that AI’s role in education is expanding faster than the data needed to track it, which means the 6 percent figure is the best available reading of a situation nobody is yet measuring properly.