The research brief.
A growing body of cited research on what AI is actually doing inside K–12 classrooms — what students do with it, what faculty observe, and what the cognitive baseline already looked like before these tools arrived.
Student-AI activity is overwhelmingly about skipping the work, not understanding it.
Education Week, working with Securly's school-network monitoring, analyzed 1.2 million student-AI sessions across more than 1,300 districts in early 2026. Across that real-world dataset, 95% of interactions were students trying to have the AI complete their assigned work entirely — not asking for help understanding it.
This is the first dataset of its size to look at what students actually do with AI on school networks, rather than what they self-report. The pattern is consistent across grade levels and subjects: AI is being used to bypass the thinking, not to support it.
Education Week · Securly · March 2026College faculty say AI is directly reducing student originality.
In a February 2026 College Board survey of college and university faculty across the United States, 84% reported that AI is directly reducing student originality and deep engagement with the material. This is near-universal agreement among the people who actually grade the work.
Faculty are positioned to see the change earliest: they read the same student writing over multiple terms and notice when reasoning patterns flatten. Their concern is specifically that work submitted today looks more polished but reflects less thinking than it did even a year ago.
College Board · February 2026 ↗Students themselves now say AI is eroding their critical thinking.
RAND Corporation's March 2026 research on student AI use found that 67% of students now admit, in their own words, that using AI for schoolwork is actively eroding their critical thinking skills. That number was 54% earlier in the same school year.
The students see it happening before their administrators do. This is one of the clearest patterns in the research: the people most exposed to AI's substitution effect are reporting it first.
RAND Corporation · March 2026 ↗AI use for homework crossed into majority among students — in eight months.
From May 2025 to December 2025, the share of students using AI for homework rose from 48% to 62%. This is the fastest adoption of a cognitive-dependency tool in recorded educational history.
The jump was driven almost entirely by middle and high schoolers — the age range where habits of mind around effortful learning are still forming. Adoption that fast, in that age range, with that little institutional response, is a category of risk schools have not faced before.
RAND Corporation · March 2026 ↗U.S. nine-year-olds posted their first-ever math score decline — and the largest reading drop since 1990.
Before generative AI became a classroom default, U.S. nine-year-olds recorded their first-ever math score decline on the National Assessment of Educational Progress, dropping 7 points between 2020 and 2022. Reading scores fell 5 points — the largest reading decline since 1990.
AI did not create a learning crisis. It arrived during one. Any school's AI policy has to be read against this baseline: students were already losing ground on foundational cognitive work before AI made shortcuts trivial.
National Center for Education Statistics · NAEP 2022 Long-Term Trend AssessmentThe steepest math decline the OECD has ever recorded.
PISA 2022 — the OECD's tri-annual assessment of 15-year-olds across 81 economies — recorded the steepest mathematics decline in the program's history: a 15-point drop since 2018. Reading declined 10 points; science 4.
The decline was unprecedented in scale and broad-based across systems. PISA's own analysis ruled out the pandemic as a sufficient explanation; the drop began earlier and continued beyond reopenings. The cognitive baseline was already weakening before AI entered the picture.
OECD · PISA 2022 Results · December 2023One shape across the research.
Students are using AI overwhelmingly to skip the thinking, not support it. Faculty closest to the work see originality declining. Students themselves are starting to report the cognitive cost. Adoption is accelerating into the age range where the cognitive habits matter most. And the foundational baseline — what schools could count on students arriving with — was already deteriorating before AI showed up.
Most of the public conversation about AI in schools is still about whether to allow it. The research is past that question. AI is in your classrooms. The question is what kind of AI environment the school is willing to take responsibility for.
Or read about our approach: How it works →