Not long ago, the question about AI in schools was how fast to bring it in. Now a growing number of states and districts are asking how to hold it back. Newsweek mapped where that is happening, and the University flagged the piece because research from its own College of Education and Human Development (CEHD) appears in it.
The strictest move on the map is New York City's. For the 2026-27 school year, it has paused student-facing generative AI from pre-K through 8th grade and banned AI chatbots in every grade, though some limited high school pilots continue. In Albany, Assemblymember Robert Carroll is sponsoring Assembly Bill A9190, which would bar most classroom AI before 9th grade, with limited exceptions. Los Angeles Unified has restricted students' use of generative AI on school devices. California's Department of Education has issued guidance on safe use, not a statewide ban.
Most other states are writing rules rather than bans. Oklahoma's Responsible Technology in Schools Act, signed in May 2026, requires districts to have written policies by 2027-28 and to disclose the AI tools they use and the data those tools collect. It also lets parents opt their kids out without any academic penalty. Maryland's Artificial Intelligence Ready Schools Act requires statewide guidance, district policies, AI coordinators and teacher training. Idaho told its education department to build a framework covering privacy, procurement and academic integrity. Virginia started a pilot program to test AI in classrooms. Alabama and Utah went the other way and added AI literacy to required coursework. Minnesota doesn't appear on the map.
Here's where UMN comes in. Newsweek cited an October 2026 expert-consensus report from CEHD on how generative AI can help and hurt learning from kindergarten through graduate school. Panayiota Kendeou of CEHD is one of its 3 lead authors, along with Jeffrey A. Greene of UNC-Chapel Hill and Nia Nixon of UC-Irvine. A panel of 30 learning scientists and education researchers rated 65 learning processes using a structured consensus method, and they reached consensus on 42 of them.
The warnings are broad. The panel found memory, critical thinking, creativity, self-regulation, motivation and persistence all vulnerable. Writing was the only subject-specific area where the experts agreed on a risk. They also flagged collaboration, belonging and social-emotional development. AI looked most useful for study strategies: learning from feedback, self-testing, tutoring and deliberate practice.
The authors call the pattern a "GenAI paradox." AI hurts students most when they first take in new material, helps them most as they lock it in, and becomes a risk again when they move to higher-order thinking. So the advice is to learn the material without AI first, use AI to reinforce it, and think on your own when applying it to something new. The report also admits a weakness. Experts rated these processes as highly important, but the evidence behind them was consistently weak. The authors sum it up as "big feelings, little evidence." Lawmakers drafting AI rules might want to keep that line in mind.
If you want to dig in, the full report and the ratings for all 65 processes are on CEHD's report page.
