Somewhere in the gap between what a classroom does and what a school district can explain about it, an entire generation is learning to work with artificial intelligence. A national survey released on September 2 puts numbers to that gap, and the numbers are less flattering than the adoption headlines suggest.
What Happened
IBM published new U.S. research on artificial intelligence readiness in K-12 schools, conducted by Morning Consult among 2,048 American adults in July 2026. The sample split almost evenly between 1,019 K-12 education professionals and 1,029 parents of school-age children, with a margin of error of roughly three percentage points on each side. Alongside the findings, the company announced the IBM K-12 AI leaders fellowship, a professional learning program built for superintendents, district administrators and classroom educators.
The adoption figures are the part everyone will quote. Seventy-six percent of middle school educators and 73 percent of high school educators say AI is used in their classroom at least weekly. Among elementary educators the figure drops to 45 percent, which is still nearly half of the youngest grades. Nearly half of high school teachers — 45 percent — report AI showing up daily or almost daily. Whatever the policy debates of the last three years produced, the tools arrived first and the frameworks are still in transit.
The preparation figures run in the opposite direction. Only 20 percent of K-12 educators say they have received extensive AI training. When asked what most blocks their ability to teach AI literacy, 42 percent named a lack of training or professional development, and 34 percent pointed to thin curriculum and instructional materials. A teacher using a tool weekly without having been trained on it is not an unusual situation in American education, but the scale here is new.
Confidence in the system as a whole is low on both sides of the desk. Just 24 percent of educators and 16 percent of parents say the U.S. education system is adapting very well to advances in AI. IBM's inaugural fellowship cohort launches this fall with up to 100 superintendents and education leaders from the greater New York area, with credentials delivered through the company's free SkillsBuild program and a curriculum built around capstone projects.
Why It Matters
Technology adoption curves in schools usually look nothing like this. Interactive whiteboards took the better part of a decade to reach saturation. One-to-one laptop programs moved district by district over years, gated by bond measures and procurement cycles. Generative AI skipped all of it, because it arrived through the browser and the phone rather than through the purchase order. Nobody had to approve it for it to be in the room.
.jpg){kind=link}
That is why the training number matters more than the adoption number. Eighty percent of educators are working with a technology they have not been extensively trained on, in a domain where the failure modes are subtle. A model that fabricates a citation does not announce itself. A student who leans on AI for the reasoning rather than the drafting does not look different on the page. Detecting either requires exactly the kind of professional judgment that extensive training is supposed to build, and most teachers report not having received it.
The curriculum gap compounds this. When 34 percent of educators cite missing instructional materials, they are describing a vacuum that gets filled by whatever is available — vendor tutorials, social media threads, district memos written by people who are themselves improvising. IBM's fellowship is one attempt to route around that, and it is worth naming the obvious: a technology company offering the training playbook for how schools should teach about technology is a structure with its own incentives, however useful the content. The OECD's 2026 Digital Education Outlook makes the same point about the broader ecosystem, where private providers have moved faster than public curriculum bodies.
There is also a workforce signal buried in the data. Forty-two percent of K-12 educators overall believe students are prepared for jobs in an AI-driven economy, but that rises to 56 percent among middle and high school educators, and 56 to 59 percent say students are ready to think critically about AI outputs. Educators who see students working with these tools daily are more optimistic than those who do not — which could mean familiarity breeds confidence, or that the confidence is calibrated to what they actually observe. The survey cannot distinguish between the two.
Reaction
The sharpest split in the study is not between optimists and skeptics. It is between the people in the classroom and the people paying for it.
Seventy-seven percent of parents say they want input on how AI is used in their child's classroom. Only 20 percent say they clearly understand the AI guidance their child actually receives at school. That is a 57-point gap between wanting a voice and having enough information to use one, and it explains why the fixes parents ask for are so procedural: 51 percent want clearer guidelines, 48 percent want greater transparency about classroom AI use. Neither request is about the technology. Both are about being told what is happening.
Educators and parents also disagree on whether AI belongs in school at all. Fifty-nine percent of educators think AI literacy should be required or offered as an elective, against 40 percent of parents. Parents are more than twice as likely as educators to say AI has no place in K-12 education right now — 28 percent versus 12 percent. Where they agree is on timing: among those who see a role for it, 41 percent of both groups name middle school, grades 6 through 8, as the right starting point.
Within the educator sample, the concerns sort by role. Classroom teachers rank student dependency on AI at 52 percent and cheating or plagiarism at 47 percent as their leading worries — the daily texture of the job. Administrators are more likely to cite insufficient training, 26 percent against teachers' 19 percent, and the cost of AI tools, 22 percent against 11 percent. Teachers are watching students. Administrators are watching budgets. Both are correct about their own vantage point.
What's Next
IBM's fellowship starts small and regional: up to 100 leaders from the greater New York area this fall, with plans to expand to additional states within the first year through what the company describes as a scalable cohort model. Reaching hundreds of education leaders is a meaningful pilot and a rounding error against the roughly 13,000 school districts in the United States. The question the program has to answer is not whether 100 superintendents find it useful. It is whether what they build travels.
The regulatory picture is moving in parallel. New York City's Department of Education has issued guidance requiring AI tools to pass a bias and equity review before deployment across its 1.1 million-student system. That is one district, but it is the largest in the country, and procurement rules at that scale tend to set de facto national standards for vendors who cannot afford to build a separate product for New York. Districts watching that process will have a template regardless of whether they write their own.
The other thing to watch is the higher education signal, which is genuinely unsettled. Forty-one percent of educators say AI has made a four-year degree feel more essential. Parents lean the other way: 30 percent say it has made a degree feel less essential, against 24 percent who say more. If that divergence holds across a few more survey cycles, it will show up in enrollment before it shows up in policy, and the people who track admissions will notice before the people who write curriculum do.
Closing Thoughts
The most quietly damning number in the study is the one about parents who do not understand the guidance their child receives. It is not a finding about artificial intelligence. It is a finding about communication, and it would look almost identical in a survey about grading policy or standardized testing. What AI has done is compress the timeline so that the usual lag between practice and explanation is now visible to everyone at once.
There is a version of this story where the training gap closes on its own, the way it eventually did for word processors and search engines and learning management systems. Teachers figure it out, informal norms harden into departmental practice, and in six years nobody remembers what the fuss was about. That version is plausible. It is also the version where the norms get set by whoever is loudest in the first two years, which is not obviously the same as whoever is most thoughtful.
The alternative requires something less organic — deliberate professional development, curriculum written by people who understand both the pedagogy and the failure modes, and district guidance clear enough that a parent can read it and know what it means. IBM's fellowship is a bet on that second path. Whether it is the right bet depends less on the quality of the curriculum than on whether a hundred superintendents in New York can turn what they learn into something that survives the trip back to their districts. Most professional development does not. This one has better odds than most, and that is a lower bar than it sounds like.
한글 요약
IBM이 9월 2일 발표한 미국 K-12 학교의 AI 준비도 조사에 따르면, 교실에서의 AI 사용은 이미 일상이 됐지만 이를 뒷받침할 교사 교육과 학교 정책은 한참 뒤처져 있습니다. 조사는 모닝컨설트가 7월에 미국 성인 2,048명(교육 종사자 1,019명, 학부모 1,029명)을 대상으로 실시했습니다. 중학교 교사의 76%, 고등학교 교사의 73%가 주 1회 이상 교실에서 AI가 사용된다고 답했고, 고등학교 교사의 45%는 거의 매일 사용된다고 응답했습니다.
반면 충분한 AI 교육을 받았다고 답한 교사는 20%에 그쳤습니다. AI 리터러시 교육의 가장 큰 걸림돌로는 연수 부족(42%)과 교육과정·교재 부족(34%)이 꼽혔습니다. 미국 교육 시스템이 AI 변화에 잘 대응하고 있다고 본 응답자는 교사 24%, 학부모 16%에 불과했습니다. 학부모의 77%는 자녀 교실의 AI 사용에 의견을 내고 싶다고 했지만, 학교의 AI 지침을 명확히 이해한다고 답한 비율은 20%에 머물렀습니다.
IBM은 조사 결과와 함께 교육감·학교 관리자·교사를 위한 'K-12 AI 리더스 펠로십'을 함께 발표했습니다. 올가을 뉴욕 광역권에서 최대 100명 규모의 첫 기수가 시작되며, 무료 교육 프로그램인 SkillsBuild를 통한 인증과 캡스톤 프로젝트 중심 커리큘럼이 제공됩니다. 미국 전역의 학구가 약 1만 3천 곳인 점을 고려하면 시범 사업 성격이 강하지만, 뉴욕시 교육청이 AI 도구 도입 전 편향성·형평성 심사를 의무화하는 등 제도 정비도 함께 움직이고 있어 향후 확산 방식이 주목됩니다.