AI is already coming for the classroom

That gap between the headline and the reality is, in miniature, the story of AI in K-12 education right now.

American schools are under pressure from every direction. Test scores still haven't fully recovered from the pandemic. Teachers are leaving the profession. Budgets are tight and getting tighter. Into that context arrives a technology that promises to save time, personalize learning, and prepare students for an economy that will supposedly demand AI fluency. It's not hard to see why schools are reaching for it.

A thousand different experiments

Pick a country, almost any country, and you will find a school system experimenting with AI. In Kazakhstan, a financial services company struck a deal with OpenAI to bring ChatGPT to 165,000 educators. El Salvador announced a partnership with Elon Musk's xAI to build an AI tutoring system for more than a million students.

Iceland, by contrast, is running a cautious national pilot in which several hundred teachers are experimenting with AI for lesson planning while students aren't involved at all, out of concern that overreliance could hollow out the learning process. Estonia has gone further, building a national AI literacy program that specifically modified ChatGPT so it responds to student queries with questions rather than answers.

In the United States, Microsoft $MSFT and OpenAI have collectively committed tens of millions of dollars to teacher training through the country's two largest teachers unions. In Florida alone, more than 100,000 high schoolers now have access to Google $GOOGL's Gemini chatbot through their school districts.

The Trump administration has encouraged all of this, launching a Presidential AI Challenge that invites K-12 students to build AI projects addressing community problems. But even in enthusiastic districts, students were learning about the contest two months after it launched. 

Many districts, including the two largest in California, had no plans to participate. The challenge offers cash prizes but no additional funding, meaning schools already running robust AI programs are best positioned to win.

The free AI tools most accessible to under-resourced schools also tend to be the least reliable, according to a 2025 Brookings Institution report. This may be, the report's authors suggested, the first time in education technology history that schools will have to pay more for more accurate information.

The evidence is not keeping up

It’s early days, but the research that does exist is not encouraging. A study analyzing AI-generated civics lesson plans found them heavy on rote recall and light on active, critical engagement. They also tended to leave out perspectives from historically marginalized communities. The tools weren't built with teachers in mind. They were built as general-purpose chatbots and handed to schools.

The Brookings report described a feedback loop in which students who offload thinking to AI do less of it themselves, and over time that atrophy compounds. "It's easy. You don't need to use your brain," one student told researchers.

A separate study from Microsoft and Carnegie Mellon found that popular chatbots may actively diminish critical thinking skills. AI systems are also designed to be agreeable, which turns out to be a poor model for the friction that builds social and emotional resilience.

There are genuine benefits. Teachers report real time savings. AI can help students learning a second language, support those with learning disabilities, and assist educators in tailoring instruction to students at different levels.

It's also worth noting that many of the tech executives pushing AI into classrooms have chosen low-tech or no-screen schools for their own children. Now they're offering something considerably more intensive to everyone else's.

Most deployments are outpacing the research by a wide margin. States are largely leaving districts to develop their own policies. Districts are largely leaving teachers to figure it out. And teachers, many of whom received their first real AI training at a Saturday workshop funded by the companies whose products were being demonstrated, are doing the best they can.

The Estonian model, skeptical and government-led and focused on AI literacy rather than AI adoption, has drawn interest from researchers globally. But it required national coordination, political will, and a negotiation with tech companies that most school systems aren't positioned to conduct.

Having a tool that answers your questions patiently and at your own pace has real value. But a lot of what school does isn't about getting the right answer. It's about learning to work with other people, sit with disagreement, and figure things out in community. That's harder to automate, and probably shouldn't be automated at all.