The story about AI in schools can be written in a way that makes sense. A student in a rural school gets help right away. It saves a teacher three hours of work to grade. Someone who is having trouble reading finally finds a tool that works for them. You’re telling the truth, but that version is just a small part of the bigger picture.
More than 500 students, teachers, parents, and technologists from 50 countries took part in a year-long global study from the Brookings Institution. The study found that at this point in its development, the risks of generative AI in children’s education outweigh its benefits. This should make every school district that has already signed an EdTech contract think twice. Not just in theory. Not some time in the future. The way things are going in classrooms and on students’ phones right now, at midnight.
The study didn’t just jump to this conclusion. Along with their fieldwork, researchers looked at more than 400 studies and held a Delphi panel. Not because AI is bad for learning in and of itself, but because the risks are built in a way that makes them different from the benefits. The benefits are usually clear, easy to measure, and simple to put on a slide deck. It’s not as dangerous. They hurt a student’s mental strength in ways that don’t show up on a test until they’re already gone, like their ability to solve problems on their own and deal with uncertainty.
To its credit, the EdTech industry seems to have gotten really good at one thing: making AI tools seem useful in the short term. Someone types in a question, gets a clear answer, and then moves on. Worked well. Clean. But teachers and researchers in this study kept bringing up a tougher question: what happens to the student who never had to think about the question long enough to learn something? The report says that overreliance isn’t just a bad habit for one person. It might be quietly changing the way a whole generation deals with problems.
What’s going on outside of school might be the part of the story that doesn’t get enough attention. In the countries that were polled, students weren’t just using generative AI in settings where they were guided and watched. They used it a lot at home, late at night, without help or any real safety measures. Teachers in Europe and other places have been talking about this worry for a while now—AI was already a part of students’ lives before it was talked about in policy circles. There is a big difference between adoption and governance.
The fact that the Brookings report doesn’t argue against AI in education makes this a really bad time for the EdTech industry. It’s calling for a higher standard: AI tools should be co-designed with teachers and communities, meant to teach rather than just tell, and controlled by rules that protect kids’ privacy and mental health. That is harder to sell than a freemium license and a dashboard full of metrics on how engaged people are.
The three main points of the report—Prosper, Prepare, and Protect—make sense. When you say them out loud, they might even seem clear. Before putting the tools to use, prepare the teachers. Real governing structures can help keep students safe. Use teaching methods that put students’ judgment first to help them do well. This is not radical in any way. Telling is how rarely it actually takes place.
It’s hard not to notice that the education sector is being asked to make long-term decisions about kids’ cognitive and social development while being under a lot of commercial pressure and with little evidence. That’s not just true for AI; schools have been lied to before. But because of how fast this wave is moving and how much is at stake, the lack of caution right now feels like more than just a policy lag. There are times when it looks like a choice.
