AI firms warn bans on AI and screens could hurt vulnerable students

The children of New York City’s public schools returned this fall with far less artificial intelligence (AI) within reach.

On September 2, Mayor Zohran Mamdani and New York City Public Schools Chancellor Kamar Samuels announced a one-year moratorium that blocks student-facing generative AI, including AI tutors and chatbots, for roughly 600,000 students from pre-K through eighth grade, about two-thirds of the largest school system in the country. Companion chatbots are barred at every grade level, and high schools keep only a narrow opening: a few thousand students in vetted pilot programs.

“Children need teachers and human connection in order to learn and grow,” Mamdani argued. It is a sweeping action built on growing concern across the country about the use of AI in classrooms.

In May, the American Federation of Teachers stood up at the National Press Club to call for banning screen time in the earliest grades, ending student-facing AI in elementary schools, and prohibiting companion chatbots for anyone under 16 years of age.

Randi Weingarten, the federation’s president, isn’t calling for a Chromebook bonfire. She cited a survey of 3,000 teachers in which 88% said their students’ attention spans were shrinking, along with reading and math scores that climbed for years before falling as classroom tech use grew.

What she wanted, she said, was the balance set right.

Across the country, at least 17 states are weighing bills to slow technology use during school hours, and other major cities like Los Angeles have resolved to implement screen time limits as well.

Some edtech companies, however, worry that blanket bans may inadvertently lead to removing access to technology tools for the most vulnerable students.

For many kids, losing access to a screen or chatbot is a minor inconvenience. For some, they argue, it’s a bite out of their overall education, with nothing to fill the gap.

For most kids, losing access to a screen or a chatbot is a minor inconvenience. For some of them, according to some people in the AI space, it’s a genuine loss with nothing to fill the gap.

Ivan Crewkov says he saw this first hand when his family immigrated to the United States. After arriving in California, his youngest daughter was learning English in a school district he describes as underfunded, and he noticed how much of her learning progress came from afterschool teaching at home.

“We noticed very quickly how much the extra practice at home mattered,” he told Unite AI. He also understood that some children in his daughter’s class may not have that kind of support at home.

Crewkov said he set out to build a company that could help fill that language learning gap. He started Buddy.ai, a COPPA-certified conversational AI tutor for language learning. He argues that, “In a well-resourced district, a blanket ban is an inconvenience. In an underfunded one, it removes something those children have no alternative for.”

In California, dual-language learners make up half of the state’s preschoolers, most from low-income households, and the state has struggled for years to hire enough bilingual teachers to serve them.

Tutoring is one of the most compelling solutions to fill holes in traditional education. A meta-analysis of dozens of randomized experiments found that tutoring has “substantial positive impacts” on learning outcomes, particularly in early grades, and especially if the tutor is a teacher.

The problem, however, is supply. Roughly 411,500 teaching positions nationwide are either unfilled or staffed by teachers without full certification, about one in eight of all positions, affecting more than 6 million students, according to a 2025 Learning Policy Institute analysis.

One teacher cannot deliver one-to-one attention to 30 children at once.

Crewkov argues that a purpose-built AI tutor can extend that teacher’s reach rather than replace her, adding repetition and feedback after the school day ends.

“It figures out exactly where you are and what you need next, updated in real time,” he said. “AI should also free teachers up to focus on what only a human can do: the hard question, the complex concept, the moment that actually changes how a student thinks.”

Not All AI in a Classroom Is the Same

For builders of AI tools in the education space, some believe that a blanket ban on chatbots could actually do more harm than good.

Abhijeet Sethi, who heads edtech strategy at LearningMate, an AI platform for education, draws the line at design intent.

There’s “a world of difference,” he said, between a general-purpose chatbot and AI “purpose-built for education — designed around pedagogy, safety, and the curriculum a student is actually working through.”

Bob Chopra, the young founder of AI learning platform IvySchool.ai, echoes this sentiment. “Every student gets a completely different experience based on who they are: different pace, different content, different examples,” he said. “Not what the average student needs, what that specific student needs.”

Crewkov added that “General tools like ChatGPT or Perplexity weren’t built with children in mind, not from a safety perspective and not from a learning one.” AI apps designed for young children should have these safeguards in place.

For its part, Buddy is KidSafe and COPPA certified and built its full stack AI technology from the ground up, without using outside LLMs like ChatGPT, Claude, or Gemini, he clarified.

John Kobs, CEO and co-founder of Giant, a COPPA-compliant AI storytelling app for kids, draws a similar line. “A child trapped in an infinite feed is consuming,” he said. “A child using AI to bring an idea to life is creating. Those are not the same thing.”

Some of the push to ban devices outright is framed as a cybersecurity and student-safety measure, not just a pedagogical one.

Charlie Sander is CEO of ManagedMethods, whose software helps school technology teams keep students and data safe on the internet. He argues that the movement to restrict technology in the classroom isn’t entirely unwarranted, but is “a bit overblown.”

He thinks the focus on the technology or device as being the problem is misguided. Rather the focus should be on how it’s used and how adults can better control the use.

“Today’s biggest cybersecurity threats aren’t tied to the device itself,” Sander said. “They’re tied to cloud accounts and the data they contain.”

Intention, Not Prohibition

This year’s Consortium for School Networking conference, the professional body for the technology directors who run K-12 districts’ devices and networks, argued that the rush to ban screens has mostly failed to separate the tools that help students from the products that distract them.

Members argued the answer wasn’t to prohibit technology, rather to better determine its intended use.

Fei-Fei Li, the Stanford computer scientist often called the “godmother of AI,” made a similar point on the Huberman Lab podcast in August. “Both things worry me,” she said. “Either denying the tool or taking away agency and motivation.”

Sander said the answer isn’t fewer devices or banning AI. “It’s creating a governance framework that makes approved tools the easiest choice while maintaining visibility into school accounts, cloud activity, and third-party applications. The districts that are most successful aren’t trying to eliminate technology—they’re managing it in a way that supports learning while reducing risk.”

Kobs said the target should be behavior, not conversation. “We should be tougher on AI products built for children, but the rules should target actual harm,” he said. Products that “encourage emotional dependency, ask children to keep secrets, isolate them from adults, or simulate exclusive relationships” should face strict restrictions, not conversation itself. “Regulate the behaviors that make AI dangerous. Don’t ban the conversation that makes it useful.”