The body regulating Florida public schools and state colleges Wednesday approved what the Florida Department of Education calls “one of the most comprehensive, parent-centered AI frameworks in the country.”

The State Board of Education  require public school districts and charter school governing boards to amend their internet safety policies to include specific guardrails and directives related to AI by July 1. 

“We are neither going to one extreme, i.e. banning AI, nor are we going to the other extreme, which is just unfettered access,” Education Commissioner Henry Mack said during the board’s meeting in Winter Haven.

Districts and charter schools, at a minimum, must make policies to “protect students, parents, and instructional personnel from unlawful, deceptive, unsafe, or unauthorized uses of artificial intelligence systems, services, applications, and tools used in connection with educational programs and operations in Florida public schools,” according to the public school rule. 

The idea, the rule states, is to “ensure that any approved use of artificial intelligence protects education records and student information, while supporting student achievement, civic literacy, academic excellence, and workforce preparedness.”

More specifically, schools will have to notify parents when teachers approve an AI instructional tool for use, including the name of the platform and classes in which it would be used and the nature of student interaction.

The policy must include a process for parents to object to AI use and provide alternatives to the tool.

“Florida will not outsource childhood to a chatbot,” Gov. Ron DeSantis said in a news release following the meeting. “We welcome tools that help teachers teach and students learn. We will not accept tools that hide from parents, harvest student data, grade kids in secret, or pretend to be a friend. In this state, a teacher stands in front of the class, a parent has the last word, and a student still has to think.”

Schools will be required to conduct additional reviews for AI tools used in PreK-5 to ensure they are age-appropriate. 

The rule looks to protect the teaching profession from being replaced by computers. It calls to “ensure that the use of AI instructional tools is limited to supplementing, and not replacing, the professional judgment of teachers and the primary right of parents to direct the upbringing and education of their children.”

Further, the policy must ensure AI is “accurate, transparent, viewpoint neutral, accessible to students and educators with disabilities.”

State colleges

The board approved a similar for the state college system, directing the 28 boards of trustees to create policies “that provide for the use and limitations of use of artificial intelligence tools in order to ensure quality of education for students, efficiency of operations, and a safe learning environment for the whole college community.”

“The rule that you accepted is not a prohibition, it empowers faculty to decide what to do, it allows the instructor to explicitly permit it or prohibit it with respect to their assignments,” Mack said of the college-level policy.

“I just think it’s a great thing that we put the guardrails, but the guardrails are so nuanced that it allows for flexibility and innovation and faculty ownership in a way that does not compromise or usurp the faculty’s ability to teach what they want.”

This story was originally published by Florida Phoenix.

is part of States Newsroom, a nonprofit news network supported by grants and a coalition of donors as a 501c(3) public charity. Florida Phoenix maintains editorial independence. Contact Editor Michael Moline for questions: info@floridaphoenix.com.

Co-founder and CEO of , a Nashville-based nonprofit that works with teachers to provide high-quality resources and instructional support, she got curious about AI after seeing the tool in action. 

It’s designed to help students read texts closely, and it did it in a way that was “very different than any tech I had seen,” Freitag recalled. It made her think that AI “may be a horse of a different color. It may be really, actually a step forward in capabilities.” But she wanted to look past the hype. 

A former assistant commissioner of curriculum and instruction in Tennessee, Freitag got the idea to take a closer look at the full spectrum of AI tools. “I wanted to see it as a mom. I wanted to see it as an educator, and I wanted to see it to be a good partner to our schools.”

So she and a colleague last January began a six-month investigation that would eventually take them to 16 school systems in seven states. They sat with students as they worked through AI-enabled lessons, watched teachers teach with AI, and interviewed more than 100 teachers, school system leaders and ed-tech CEOs.

The results of their first foray are : Instruction Partners on Monday released the initial findings in what it is calling its AI in Action Learning Tour — a year-long effort to observe how actual students and teachers use AI. 

The report looks at 20 student-facing AI products already in classrooms, broken down into three categories: general-purpose chatbots such as ChatGPT, Claude and Gemini; education-focused platforms such as Brisk Teaching, Google Classroom, MagicSchool and Playlab, and targeted instructional tools built around a specific instructional jobs — they include Amira Learning, Khanmigo and Quill.

“AI is definitely not a single category,” Freitag said. “We saw a huge range of products designed to do a huge range of things.”

They framed their inquiry around two questions: 

1. How does it compare to paper-pencil instruction?
2. How does it compare to the best possible learning experience in that subject?

Freitag and her colleague found a decidedly mixed bag, with a few key surprises.

“I saw tools that held a higher bar for learning than we see in paper pencil instruction and did more to tailor the questions and the feedback for kids to the specific sort of responses they were giving than a teacher could possibly do all at the same time,” she said. “So I really walk away thinking there’s a lot of good here that we need to leverage — and a lot of bad here we need to mitigate.”

They found that general-purpose chatbots like Claude and ChatGPT pose the biggest threat to student thinking, while the platforms — Brisk, MagicSchool and the like — produced different results, depending on how teachers used them.

The targeted tools, such as Khan Academy’s Khanmigo and Quill, a writing app, produced the most consistent and strongest learning experiences, they found.

And teachers told them they’re frustrated by apps kids can easily game — in one case, students found that by simply hitting the “Refresh” button on a computer, they could complete an assignment by answering the same question repeatedly instead of the lesson’s full list of questions. Many students also learned that they could simply Google an answer and paste it into their lessons. 

In another case, an app gave students detailed feedback on their writing — “I thought the feedback was really good and useful and actionable,” Freitag said. But she watched in horror as students simply ignored it and began rewriting without consulting it.

In another instance, she sat with a student as he worked on writing a book based on characters he liked, focused on letter sounds he needed to practice. “He created the book, and he said, ‘Now it’s your turn. Let’s create a book for you.’ And I was like, ‘But are we going to read your book?’ And I realized he was just making books without reading them.” 

The boy, she said, simply liked making the books.

“As someone trying to make sense of where AI has a role to play or not, I think seeing how kids are actually using products matters much more than understanding what they’re designed to do,” Freitag said. “And there are many cases where what I saw in action was quite different than what I expected to see based on the product demo.”

In fact, many of the apps that give students the ability to work together shone in Freitag’s review. By contrast, several apps designed for solo use turned out to be “boring” to many students, who spent perhaps 10 minutes engaged. “By minute 42 of the lesson, they are just tuned out.”

In many cases, she said, the same product produced dramatically different results — and gave kids different experiences — depending on whether teachers built in clear routines and actively monitored dashboards in real time.

In one instance, a teacher taught a lesson with a math app that allowed her to display four students’ solutions to a problem side-by-side on a Smartboard, allowing students to evaluate the solutions. “I was stunned by how much kids learned from looking at each other’s work in that way,” Freitag said. Even with just a little feedback, it was effective. “When it was used socially, it really created a ton of learning opportunity for kids.”

The apps that worked did something that teachers couldn’t do alone, such as offering instant feedback or a quick analysis of an entire group’s work. Those, she said, “were the products where I was like, ‘Wow, this is extending what a teacher could do.’”

But the report emphasizes that AI shouldn’t cut teachers out of the equation. They still have a vital role to play and AI should never be seen as replacing teachers. “My take is that the future is hybrid,” Freitag said. “Teachers have a key and lead role, and there’s tech that gives them an assist.”

In fact, one of the most promising AI innovations is its ability to identify where students are struggling with foundational skills — and where they need more practice. But forcing students to practice those skills in isolation could be counterproductive. “If we help kids practice a bunch of reading skills without actually paying attention to the text or the knowledge that they need to access that text,” Freitag said, “we’re not going to see reading scores improve.”

She also worries about AI’s near-miraculous ability to re-level reading passages with a single click, one of the most popular features of most reading tools. While it can help slower readers understand a text, that releveling can become “a permanent lower setting” rather than a bridge back to grade level.

Arman Jaffer, CEO of Brisk Teaching, one of the apps reviewed, said his philosophy is that releveling “should be used to increase entry points to high-quality instructional material,” but not to limit how far students can go.

“AI is something that’s incredibly powerful that can help you do things like relevel,” he said. “You can make the entire internet accessible to your students, but at the same time, a teacher who might not have the best instructional practices might be using it to hold back students.”

Jaffer said the findings underscore the need for coaching and professional development for teachers as AI emerges in the classroom. “I think every AI company, every technology company, every vendor, truly needs to be a partner to their district or their school or their organization.”

Peter Gault, CEO of Quill, an AI-powered writing app that was also reviewed, said Freitag and her partner watched students using the app in New Jersey and worked hard to unpack how it worked. “I was really impressed with how much due diligence they did. They really went under the hood to understand what works and why it works.”

The duo, he said “did the most thorough deep dive of anyone who has looked at our AI.”

Alex Sarlin, co-host of the podcast, called the new report “amazing work,” noting that Freitag spoke not just with teachers and students but with tech CEOs. “She’s talking to all sides of the equation,” he said. 

The time she and her colleague spent in classrooms is key, Sarlin said. “That gives a much richer picture of how AI is actually being used in the classroom than we usually see, and certainly more than we see in mainstream coverage, which tends to be very surface level and a little panicky.”

This fall, Freitag’s team has grown to three people — they’ll soon look at 30 more products and report on them this winter. And Instruction Partners plans to work with Stanford’s AI in Education Lab to independently vet impact claims by winter 2027. The new report notes that just seven of the 20 products have independent studies examining outcomes across student subgroups.

For her part, Freitag said teachers are hungry for clear guidance around AI. “We work with instructional leaders across the country, and they said, ‘We are getting inundated with sales pitches, and we are getting inundated with calls now to “Ban it all,” and we don’t really understand where there’s value in this.’ So that helped me see that the debate about AI in schools is moving faster than real information.”

More than 139,000 students ages 10 to 13 from all backgrounds and 3,100 educators from 126 cities in 39 countries have connected via e-classrooms over the past decade. 

I am one of them.

Participating in a global digital exchange program helped foster a deep curiosity about the world and made me realize that it is a bigger, more connected place. It inspired me to serve on the University of HawaiÊ»i Board of Regents and to seek out programs and opportunities like the , which centers intergenerational collaboration to move toward a sustainable future. I have kept chasing that same feeling of being genuinely changed by someone else’s reality.

As a student at the University of HawaiÊ»i at Mānoa, I think often about what education is actually for and what it represents. Right now, everywhere I turn, the conversation seems singularly focused on one question: “What do we do about artificial intelligence?” The debates around artificial intelligence in education usually center on whether students are using AI to cheat or as a learning resource. Questions like: Can students use AI on assignments? Will AI replace teachers? How do we design curriculum for a post-ChatGPT world? 

These are real concerns, but in our urgency to respond, we’re neglecting something equally important that AI can’t replicate: human connection. 

Being able to talk with a student across the ocean whose daily life looked nothing like mine, or hearing another student’s perspectives on climate and the future, created a human connection that AI could never replicate. That connection required me to listen, to challenge my beliefs and to update my thinking in real time — not because I’m asking a chatbot, but because a real person is trusting me with their experience. 

I was reminded how valuable this last April when I flew to Madrid to meet Global Scholars alumni from around the world. Seeing everyone in person was a completely different experience from interacting online. Shaking hands, sharing meals and being vulnerable with one another face-to-face strengthened the cross-cultural bonds we started building online. 

AI can expose you to diverse viewpoints. It can summarize them, translate them and even argue them. However, it cannot make you feel empathy for someone else’s world or help you grasp its weight. It cannot hold you accountable to another person across a cultural divide. It cannot recreate a substantive discussion about solutions to urgent global problems.

I remember my first live conference call 12 time zones away, with middle school students in Barcelona. I told them about drought concerns on Kawaihae, where it used to be green. And they told me about water restrictions in Catalonia, a place I’d never heard of. ”Do you really live near volcanoes?” someone asked me. I almost laughed, until I realized how little we actually knew about each other’s worlds. A kid from a small town on HawaiÊ»i Island had just as much to contribute to the conversation as a student in one of Europe’s great cities. 

In Hawaiʻi, human connection is embedded in our culture, in the concept of aloha, in the practice of showing up for your community. We understand that relationships are not just a part of learning; they are the learning itself. Developing cross-cultural empathy comes from building strong relationships and having meaningful conversations that challenge your own beliefs.

I sometimes think about that classmate in Barcelona and whether she remembers the eighth-grader from Hawaiʻi who thought volcanoes were just part of the backyard. I hope more kids get the opportunity to be surprised by each other in ways no algorithm or AI chatbot can replicate. We should be building more of those connections. Not less.

Across the country, elementary schools employ an estimated 67,000 speech language pathologists — experts on the front lines of diagnosing early language and reading struggles. 

The problem? To meet , schools of them. 

It’s the same for reading specialists, highly trained educators who have made literacy their professional calling card. There are simply out there with that expertise. 

These shortfalls are part of the reason that students’ reading proficiency is . Today, 40% of fourth graders can’t do things like provide support for ideas related to a story’s plot or characters. That’s up from 33% in 2022. A third of eighth graders can’t pick out the main idea of an informational text. 

Ensuring that every child has foundational literacy skills is going to be a Herculean task. Based on my work with researchers and technologists, I see enormous potential in automatic speech recognition as a tool for helping to make up for the shortage of specialists. After listening to children read aloud, this artificial intelligence-powered technology can help assess their progress by identifying weaknesses, provide them with feedback and give teachers and parents recommendations for interventions. The promise of this technology is substantial – if platform developers and their funders can get it right.

This is particularly true in K-2, when students are learning skills that for lifelong reading and success. By assisting speech pathologists and teachers, automatic speech recognition could make sophisticated early literacy screening more accessible and affordable — and it should only get better as AI advances.

Speech recognition technology has a , and it has been used to support literacy development in young readers since 1990. That’s when Carnegie Mellon University launched , a computer program that displayed stories on a screen and listened to children as they read the text. The program intervened when the student made a mistake, got stuck or asked for help. It also gave teachers feedback and advice on the child’s progress and struggles.

The approach worked. A compared the literacy skills of students in grades 1 to 4 who used the program with those of classmates who engaged in sustained silent reading. Those who used the technology-powered intervention, especially in first grade, outpaced their peers in word identification, comprehension, fluency, spelling and understanding of individual sounds.

Some modern versions of automatic speech recognition serve primarily as a screener, like Amplify’s , which provides teachers with immediate feedback on students’ reading fluency and specific areas where they need extra support. Other innovations use the technology to not only assess, but intervene in real time, like ’s AI reading tutor. An , the tool works in English and Spanish and serves 5 million students in all 50 states. Amira is also collaborating with the Children’s Hospital of Philadelphia and the University of Pennsylvania Data Center to develop an for neurodiverse students. 

Automatic speech recognition is particularly strong at helping students decode words, as it improves their ability to match written letters to the spoken sounds they make — a key skill in literacy development.

This spring, Renaissance Philanthropy launched with one moonshot goal: develop AI-powered tools that, over the next five years, will reduce the number of struggling young readers by half. 

Short for Learning Engineering Virtual Institute, LEVI provides five years of funding to researchers and developers who share their knowledge and learnings while tackling big challenges. We recently the teams selected to work toward the literacy moonshot: Reads Labs at Harvard University, OxEd, LitLab.ai and Project Read with the University of Florida Literacy Institute.

Of course, there are challenges to deploying automatic speech recognition in classrooms if it is to reach its full potential. Developers, companies and researchers need to invest in the that are needed to train the AI behind these tools. Importantly, these must include a range of children — English learners, special education students, etc. — so the apps and platforms can assist kids with diverse needs. Finally, developers must ground their tools in cognitive science on how the human brain learns, then validate their platforms through rigorous classroom testing.

Automatic speech recognition can never replace a skilled and caring teacher, but it can serve as a patient, non-judgmental expert companion for every educator trying to better understand who is struggling to read, and why. With thoughtfully designed, research-backed screeners and tutors, the nation can have a powerful new tool to tackle its literacy challenge, and give all students more of the personalized reading help they deserve. 

Disclosure: Walton Family Foundation provides financial support to LEVI Literacy and ĂÛÌÒÓ°ÊÓ.

A blank document had become intimidating, not because she had nothing to say, but because she forgot how to say it alone. 

That scene from one of our classes underscores the unique challenge our generation faces as the first to grow up with artificial intelligence being a part of everyday life. High school students across the country share the struggle of navigating education with this new technology. 

According to a , the vast majority of teens who use AI for schoolwork do so without clear guidelines from their teachers, turning generative tools into an easy shortcut rather than an educational aid simply because no one ever taught us how to use them properly. 

This isn’t the first time a young generation has grappled with new technology in everyday life. During the rise of the internet in the late 1990s, the original digital divide was a battle of infrastructure and wealth. It split our parents’ generation into those who had computers and dial-up at home, and those who were left completely offline. 

Decades later, proved that this original physical gap directly fueled a deeply polarized labor market. This divide compounded existing socioeconomic inequalities, leaving those who lacked technology access with restricted career and wage prospects. The difference between the rise of the internet and AI is that the AI divide is defined by agency — who is taught to command systems and who is left to navigate them in the dark — not access.

Some school districts have acknowledged this and have taught students how to use AI tools by practicing prompt engineering, learning to verify AI-generated facts and enhancing the skills that the workforce will demand. For instance, partnered with higher education and industry leaders to launch a district-wide AI literacy initiative. The district integrated specific frameworks into high schools to ensure students graduate with the critical thinking and ethical skills necessary to navigate an AI-driven economy.

However, in other schools, students are banned from using AI entirely and are not taught how to spot bias, talk to AI, or to use it ethically. For instance, has adopted strict frameworks that restrict major AI platforms, except Microsoft Copilot and Canva AI, across school networks — treating any unapproved use as an academic integrity violation. 

Simply warning students against using AI without teaching them how the technology works or how to question its outputs doesn’t stop them from using it. It only leaves them to use it passively, without developing critical thinking skills to optimize it. 

According to the Lumina Foundation, over half of high school students regularly use AI tools for schoolwork, even though . When policy ignores this reality, schools fail to eliminate AI use — they only succeed in eliminating AI literacy.

In addition to students, many teachers and administrators are also being left behind simply from the lack of guidance and necessary digital infrastructure. In fact, as of 2026, lack the time to develop an AI training curriculum, and 75% lack the knowledge to do so. Without a standard curriculum that simultaneously educates our teachers and students, schools risk creating a future workforce where those unfamiliar with AI tools are systematically left behind. 

Although AI currently poses challenges in schools, it can transform learning if integrated deliberately. As students from Virginia and Wisconsin, we call on state legislators and school boards to pass legislation mandating AI literacy standards tied to industry benchmarks as a graduation requirement. 

To enforce this, state boards of education must tie direct funding to AI-resilient curriculum compliance, while also allocating funds for teacher professional development so educators across all subjects — not just technology fields — are trained to teach ethical AI usage.

Educational communities do not need to start from scratch: Successful models already exist. recently implemented an initiative that involves its members working with high school teachers to develop a curriculum — based on Google’s AI Essentials online course — designed to teach high schoolers how to safely use AI.  

If we as a society don’t teach all students how to interact with AI, young people will not only be given a weaker education, they will be locked  out of the new AI-integrated economic ecosystem. Legislators, school boards and educators have a responsibility to ensure that young generations are able to use AI. 

That’s progress. But at this rate, it would take over 30 years for schools to reach the the American School Counselor Association’s (and still overwhelming) 1:250 ratio.

Students and schools aren’t waiting for human help to arrive. Nearly of students are already turning to artificial intelligence chatbots for college and career guidance, and schools are starting to use AI to answer students’ frequently asked questions to free up counselors’ time.

Turning to tech is rational when human resources are scarce. But swapping chatbots for overworked counselors masks a more fundamental problem: For decades, schools have tried to make a one-to-many advising model work.

AI could be a tool to help schools create something different: a many-to-one model in which each young person is surrounded by a strong network of teachers, alumni, family, mentors, employers, coaches and counselors working together to help students make progress.

That model is fundamentally grounded in a proven driver of student success: relationships. Research has long shown that relationships lead to success in , increase and , and ultimately unlock . 

That was the vision behind a pilot project my colleague Anna Arsenault and I led with 10 college and career guidance technology platforms, exploring how to build tools that didn’t just give AI-generated advice, but made it easier to find, reach and rely on their broader human networks. We worked with the designers of each platform to create and test features that would help students recognize their existing connections, practice reaching out and find new ones who assist them in achieving their goals.

Our new report, “,” describes three trends that highlight what schools and tools can do to move toward a many-to-one model of college and career support.

First, students need help recognizing the network they already have. This includes not only the many assets around them, but also the relationships students overlook entirely.

ESAI, a college admissions platform, piloted a network-mapping feature where students organized extracurricular activities and identified the “people in their corner” who could help them pursue their out-of-school interests. Family and teacher connections together accounted for 40% of selections; counselors made up just 10%, peers only 4%. Another platform, iCouldBe, analyzed student network maps from the past school year and found 54% of relationships were family members, 30% school-based and but just 1% came from work, internships or volunteering, school alumni, neighbors or faith communities.

Data like these don’t just diagnose a gap; they hand students and counselors a starting point to act on it, surfacing potential guides and mentors in students’ lives who would otherwise stay invisible.

Second, connection has to feel authentic. Even after students identify who’s in their corner (and who they might be missing), it’s often hard to get them to engage with those connections. The thing that mattered most: authenticity. Students didn’t want outreach to feel transactional or one-sided. 

Overcoming that involved creating content in which students talk about the benefits of networking in their own words, rather than adults explaining it to them. For example, Reach Pathways recorded videos of students describing how their relationships helped them accomplish their goals. Other platforms found it was critical not to force students to make an ask before they feel ready. For example, ESAI ‘s team discovered that after identifying “people in their corner,” over half (53%) of students first wanted to express gratitude to those supporters before asking them for something. Only 14% made an outright request for help. 

In other words, students are more likely to turn toward their networks if they see their peers doing so, and if they can start real gratitude — not with a premature request.    

Third, students need help turning relationships into sources of support. Even with more emphasis on authenticity, many students still reported that connecting with people and asking for help is daunting. Students wanted more assistance in reaching out and meeting the right people to support their college and career aspirations. 

To address that, some platforms, like Climb Together, Roadtrip Nation and FutureFit AI, built chatbots to coach students in seeking help, offering the chance to draft emails and even rehearse conversations with industry professionals. (Unlike most college and career chatbots, these tools weren’t meant to provide postsecondary advice — they were designed to build the confidence to reach a real person.) Across the board, students in the pilots using these chatbots reported modest increases in awareness of the networks around them and confidence in reaching out for help.

Other tools sought to curb anxiety by helping adults, like counselors and educators, step in to offer warm introductions. For example, Overgrad, a college and career planning platform, piloted an alumni directory to help counselors connect students with recent graduates in the colleges or industries they hoped to pursue. 

Taken together, the three trends show that a many-to-one advising model isn’t just a nice idea — it’s already taking shape with AI tools specifically designed to foster, rather than shortcut, connection. Platforms can help students see their networks, engage them without it feeling transactional and get the coaching or introduction that turns a contact into a helpful connection.

What’s possible, however, isn’t inevitable. To date, no state-level request for proposals — the documents that steer what ed tech companies actually build — lists stronger student networks as a desired outcome for college and career advising tools. Even states and districts that require postsecondary plans rarely ask whether a student can name who will help carry that plan out.

That needs to change. States and districts should demand that advising technology help students identify, strengthen and expand the human networks around them and measure whether those networks actually grow.

We could wait 30 years for the counselor ratio to reach 250:1. Or we could start building toward a different ratio altogether: many people, supporting every one student.

The 2025 administration of the Programme for International Student Assessment, or PISA, showed reading scores of America’s 15-year-olds falling by 14 points since 2022, when the exam was last conducted. Math scores fell by only two points over those three years, but have now sunk 15 points lower than in 2018. Performance in science is essentially unchanged over the last several rounds of the test.

Somewhat perversely, these deflating results mean that the United States has essentially maintained its relative positioning against the dozens of other national and regional school systems that participate in the assessment. In many of those jurisdictions, achievement in math and especially literacy has fallen even further than it has domestically. 

The ongoing slide in all but a few countries — exceptions include perennial overachievers like Singapore, Taiwan, and Japan, but also the European states of England and Estonia — indicate that the U.S. is subject to the same educational trends influencing the rest of the world. Clearly, it is not alone in suffering through a “learning recession” that began over a decade ago. Likewise, survey responses included with the PISA report show that students overseas are just as willing as their American counterparts to rely on AI chatbots: Far fewer than half of respondents across 82 participating jurisdictions said they used AI “never or almost never” for purposes related to schoolwork. 

Michael Petrilli, president of the right-leaning Thomas B. Fordham Institute, called the scores “alarming,” adding that they largely conform with those of other high-profile exams, such as the U.S.’s biennial National Assessment of Educational Progress. While academic outcomes have fallen more or less continuously for K–12 students since the Obama administration, he wrote in an email, researchers and advocates shouldn’t take solace in the equally poor performance of other countries.

“The fact that other advanced nations are also seeing declines, especially in reading, is cold comfort,” Petrilli said. “Knowledge is not a zero-sum game; we should want every child, everywhere, to achieve their full cognitive and academic potential, and that’s clearly not happening.”

Created and managed by the Paris-based Organisation for Economic Co-operation and Development, PISA made its debut in 2000. Across all students participating in the 2025 assessment, the group announced, average results in math and reading were the worst ever recorded.

Whether or not that result represents rock bottom, it follows a prolonged tumble over multiple iterations of the test. Math and reading scores have either dropped or stagnated since 2009, while reading scores have plummeted since 2012. Just in the 10 years preceding the most recent testing round, OECD reported, average reading scores fell by the equivalent of 1.5 years of classroom learning, while average math scores fell by over one year.

Vicki Phillips, head of the National Center on Education and the Economy, that most PISA participants were experiencing a pronounced diminishment of the very skills that “will increasingly shape individual opportunity and a country’s ability to innovate, adapt, and grow.”

“At exactly the moment our economies and societies are asking more of young people, too many education systems are struggling to build the knowledge and capabilities they will need,” Phillips argued.

Amid the general nosedive, America’s downturn looks less extreme. But it distinguishes itself in terms of inequality, with exceedingly large disparities separating its top-scoring and lowest-performing test takers. The “90-10 gap,” measured between struggling students at the 10th percentile and high-flyers at the 90th, grew by 50 points in reading since 2015. The gulf in science scores now stands greater than 290 scale points (out of 1,000 total), far more than any other country except Luxembourg. 

Those developments are fairly well-established, with tests such as NAEP repeatedly pointing to widening divergences over the past decade. But this year’s PISA brings some of the earliest international indicators of AI’s widespread penetration into schools. 

In a questionnaire accompanying the test, over half of all students said they enlisted chatbots like ChatGPT at least several times a month to conduct their research, write first drafts of papers, or summarize a text required for class. The percentage of students who said they seldom or never used AI for such tasks fell below 30% in every country except Japan, where 39% of respondents said as much.

Notably, too few U.S. students took part in the OECD questionnaires to reliably report their responses. But recent polling evidence strongly suggests that American pupils are highly dependent on AI to help complete their day-to-day academic assignments. 

In by the Fordham Institute, 48% of high school respondents said they regularly had AI solve math problems for them. Over one-third said they asked AI tools to summarize texts to avoid reading them, while over one-quarter said they used it for help during quizzes and tests.

Dan Goldhaber is an education professor at the University of Washington and a longtime observer of tests like PISA. In an interview, he said he worried that K–12 students would increasingly come to depend on AI for routine academic work — and, in so doing, deprive themselves of meaningful learning opportunities.

“I hire people that AI is supposed to replace,” Goldhaber said. “I worry that it makes some things so easy that people don’t learn those things, and that will lead to a huge gap in knowledge for whole cohorts of students. I’m not sure what to do about it, but I see that problem on a micro-level and think, ‘My God.’”

He warned them up front: “This is going to be the hardest class you’ve ever had.”

His students’ response left no doubt: They eagerly dug into the so-called Redhead Murders, unsolved killings of at least six young women whose bodies had been dumped along roadsides across the South from the late 1970s through early 1990s. They soon tracked down an FBI profiler, interviewed a victim’s brother, built timelines out of decades-old case files and identified several new victims. They brought in a true-crime podcaster who was already poking around the killings, developed their own profile of the killer and landed on a theory of the cases that the Tennessee Bureau of Investigation eventually supported.

Their work became the basis of their own 2024 true-crime podcast, . Later that year, Campbell gathered a new group of students to continue the work — it became the basis for an of the same name that in July cracked the streaming service’s rankings.

In an interview, Campbell said the daunting project succeeded not because his students found the work easy — just the opposite, actually. 

“I make it obvious to them up front that this is going to be very hard, but they’re going to be invested, and so that makes it easy to do hard work,” he said. “If you ask people about the most meaningful experiences of their life, they’re almost always hard.”

That insight sits at the center of a conversation now surfacing among educators. As generative AI drafts essays, solves equations and summarizes novels in seconds, a growing number of teachers say the technology has stripped away the last excuse schools have to address a problem they’ve long avoided: chronically low student motivation.

AI didn’t create the crisis, they argue. It just exposed it. In the process, the technology, which relies on random word generation, could help make deeper learning a must-have feature of K-12 education.

‘A very transactional educational process’

In interviews, Campbell and others described a kind of one-two punch that schools have withstood since the COVID pandemic in 2020 and, more recently, the advent of AI models such as ChatGPT. 

Since COVID, many students have simply concluded that school attendance is optional, said Campbell. “Why do I need to go down there when I can watch YouTube videos and get on Khan Academy and you can send [work] to me through Canvas?” he said. 

His answer: School must now offer challenging-but-doable experiences that students can’t get anywhere else.

This idea is by no means new — at other schools, students have worked with engineering firms to design houses, built a smartphone app for a children’s hospital and, in one case, helped planners rethink what services Kansas City’s updated airport needs. At one New Mexico high school, students can now earn FAA certification to operate drones designed to modernize farming techniques on their families’ ancestral land. 

Many scholars trace the movement’s roots back nearly half a century. In the mid-1980s, educator Seymour Papert, the co-founder, popularized the phrase “hard fun.” A developer of the computer language , Papert reportedly once heard a student describing how fun he found a difficult coding assignment, later , “I have no doubt that this kid called the work fun because it was hard rather than in spite of being hard.”

Then as now, the idea was something of an alien concept in most schools.

Chris Hulleman, a University of Virginia researcher and founder of the university’s , has been studying student motivation for more than 20 years — he wrote his 2007 doctoral dissertation on the topic and noted that student complaints about schoolwork’s lack of relevance go back much further. “We have just not done a great job overall in education in making school interesting and relevant to kids,” he said.

Much of that stems from a basic structure that prioritizes efficiency, usually in the form of higher scores on standardized tests. “If that’s the primary output we’re looking for,” Hulleman said, “you’re not going to have a very interesting system. And that’s sort of what we repeatedly do.”

But over the past four years, he and others said, AI has introduced its own kind of brutal efficiency, one that removes the key steps that have long forced students to at least go through the motions of an assignment. In a sense, Hulleman said, the social contract between kids and schools has changed.

“They’re questioning the ‘why,’ and then they’re looking at what they’re being asked to do,” he said. “And if it’s boring or seems stupid to them, then yeah: Why wouldn’t I use this free thing that’s at my fingertips?”

Recent research suggests that millions of students are reaching for the free thing: The College Board that the share of high school students using AI for schoolwork rose from 79% to 84% between January and May 2025, with about half using AI to brainstorm, revise essays or research topics. 

A from the University of Southern California found that most college students default to what researchers call “executive help,” using AI to get a fast answer with minimal effort, rather than “instrumental help” that deepens understanding — unless a professor actively steers them in the other direction.

Left to their own devices, researchers found, students tend to take the AI path of least resistance, even if they learn less.

Darren Lam, 18, a recent graduate of New York’s , said that while many of his classmates won’t use AI on principle, many others use it to automate school, despite concerns about academic integrity. They’re “just not even really at school anymore — basically the AI is the student.”

Even students who haven’t taken their AI use that far, he said, are still using it to do “undesirable assignments” or to get “some sort of additional time benefit that they probably wouldn’t have if they were completely honest with their work.”

Colorado math teacher Dylan Kane, who writes the Substack newsletter , said AI is automating cognitive effort at a scale far beyond earlier tools such as calculators or search engines. He used the metaphor of indoor plumbing, which freed people from hauling water, but which also erased the daily walking that once kept them healthy. As a result, people typically got less exercise. 

But he and others worry that AI might be automating cognition faster than we can find other ways to incentivize it. In the long run, he said, chatbots might be contributing to a reversal of , the decades-long rise in IQ scores generally attributed to a more cognitively demanding world. Though the phenomenon predates 2022, educators like Kane say AI is certainly worsening the slide.

“People are getting dumber,” he said, “and this is a cross-generational phenomenon.”

The response, Kane said, should be similar to the one we turned to when walking became optional: “My take is that we should just be honest about what’s happening.” 

Invoking Papert’s principle of “hard fun,” he noted, “One of the best ways to get someone hooked on exercise is to get them doing something hard, but within reach, show them the progress they make when they put effort in, and hope to create a situation where that effort is self-sustaining. We should, I think, put the same effort into ‘cognitive exercise.’”

For many students, that might still be academic learning — reading, writing and math, etc. For others, it might be chess, robotics, woodworking, debate, drama and the like. “Should those play a larger role in today’s public education?” he said. “I think there’s a good argument the answer is yes, not only as something to expose students to, but something to try and help students get good at, to experience hard fun, to show themselves what their minds are capable of.”

Rebecca Winthrop, a senior fellow at the Brookings Institution and co-author of the 2025 book , said that when tools like ChatGPT AI arrived in 2022, they came face-to-face with a school culture already ripe for change. AI quickly became layered over “what has become, sadly, a very transactional educational process.”

In a , fewer than half of middle- and high-school students said their coursework challenges them or gives them the opportunity to do what they do best. It found that just 17% “strongly agreed” with the assertion: “In the last seven days, I have learned something interesting at school.”

What’s worse, Winthrop said, AI arrived at a time when what it means to be well-educated is rapidly changing. For decades, most students could coast through school, learning basic literacy and numeracy and still find their footing in the job market. That’s no longer true. “You’ve got to have a lot of higher-order thinking skills,” she said. “The bar has shifted, and it’s much higher.”

Educator Priten Soundar-Shah, author of the recent book : How Educators Can Lead on AI and Digital Safety in K-12, said teachers can use that dynamic to their advantage. If students, for instance, scoff at foreign language instruction, insisting that their AI-powered make French class obsolete, ask them who they’d rather rely on in a pinch in Paris: a fluent French speaker or someone standing there waiting for their earbuds to process the conversation?

Students get the difference, he and others said. Perhaps as a result they’re often more conflicted about AI than adults might assume. They genuinely want to learn and are uneasy about assignments they could easily “hack,” Winthrop said, but they worry about being the only one in class to refuse a shortcut — a sort of academic in which each person thinks primarily about their own welfare.

Ben Riley, founder of and the think tank , described the dilemma more bluntly: Students, he said, “don’t want to be suckers.” If classmates are using AI to boost their grades, few teens would want to fall behind doing school the hard way.

Ask students why they go to school, said , an Ohio State University educational psychologist, and the answers typically split into two camps: some say they’re there to learn, but many say they’re there to get grades. When teachers emphasize tests and grades over mastery, he said, cheating rises; when the message shifts to learning, it falls off sharply. 

Anderman, who studies motivation and academic dishonesty, typically tells pre-service teachers, “‘A’ stands for ‘artificial.’ We made it up.” He added, “We need to do a better job of training teachers to not talk so much about the test.”

‘I deserve meaningful work’ 

While events of the past four years have forced teachers to rethink motivation, Campbell, the Tennessee teacher whose school is located near Knoxville, said the dilemma was on his mind nearly a decade ago when he conceived of the cold-case project. 

Inviting students to solve a problem that no one — not even experts — can solve sends a powerful message, he said. “That really tells kids, ‘He does believe that I can do something. He does believe I can be anything I want to be. He does believe I deserve meaningful work that has real consequences.’”

The key was in easing students into the cases’ complexity, not just dropping them into it. “I try to build them a ladder,” he said. “‘Just step up on the first rung.’”

That may be meeting with a victim’s brother to hear a family’s story. Then he invites students to meet with police and go through case files. “Before you know it, they’re building timelines,” he said. “They’re doing profiles, they’re doing geographic studies, and then they slowly begin to step up and realize they’re learning the skills, they’re meeting the people.”

But what really made the project work, he said, was its authenticity — and how it visibly helped victims’ families. 

“Kids can spot a fake in about two seconds,” said Campbell. “They know when you’re fake. They know when your problem is fake. So whenever you say, ‘Oh, this is an important problem,’ they’re like, ‘No, it’s not, because the answer’s in the back of the book. That’s stupid. Somebody already found the answer.’ ”

Nearly a decade after Campbell first brought the cold cases to students, many questions remain unanswered, foremost among them who actually did the killings. But the projects have activated his students like little else could, largely because the inquiries have given victims’ families hope for justice.

Ulcca Joshi Hansen, an education researcher, futurist and author of the 2021 book , said teachers like Campbell, who draw students into big, relatively high-stakes projects, are tracing what may be our only logical path to progress in the Age of AI. Project-based schools like , the private in Idaho, in California and those in the network nationwide, she said, take as a given that an education built around intrinsic motivation, not extrinsic markers such as grades, simply works.

Big Picture schools, for instance, are small, personalized high schools built around career-focused aspirations and a seemingly endless parade of internships with local businesses and nonprofits. And while a few students focus early on dream careers, most spend four years zig-zagging through multiple internships and experiences, often in wildly divergent fields. 

Hansen said schools can also learn from youth development programs such as and summer camps, which offer kids access to real-world, often challenging activities. “You have to start with young people being invited into their education around things that they care about,” she said. 

The obstacle, in her view, is less pedagogical than structural: our accountability system is still anchored to sit-down, written standardized tests that effectively tell schools, “You don’t know math if you can build something. You only know it if you can show it on a piece of paper.”

Educator Soundar-Shah said redesigning assessments could be a key part of solving the motivation crisis. He has moved toward what amounts to a in which students spend their out-of-class time reading books, watching video lectures or even interacting with AI chatbots role-playing as historical figures. That frees up class time for discussion, debate and writing — under his supervision. 

“I don’t do any instructional time during class,” he said. In its place, very often, are assessments in the form of in-class presentations. “There’s no point in offloading your conversations to an AI chatbot if you’re going to have to show up to class and still demonstrate that knowledge tomorrow,” he said. 

He’s also increasingly drawn to , which lets students retake assessments until they’ve actually learned the material rather than treating a single test as a final verdict.

Whatever happens, said Hansen, change is badly overdue: “For 10 years, we’ve had moment after moment after moment where the invitation to the dominant system has been actually, ‘Redesign and let go of some of the stuff you’re holding so tightly onto.’” 

While teachers like Campbell and Soundar-Shah are looking at redesigning the education system, Hansen said, most remain satisfied “tweaking and altering it to try and make it work better — and bludgeoning kids to do it better, or do more of it. And that has not worked.”

A former health industry reporter, Singer sees parallels between Silicon Valley’s campaign for coding, among other measures, and Big Pharma’s influence on medicine, noting huge school investments over the past few years by Apple, Google and Microsoft, among others.

“Never has so much market power lined up behind a single curriculum cause,” she writes.

As a result, a movement that could have helped build data science or other less glamorous, evidence-based reforms has in many cases ended up offering students “only half an education in computing, the rosy upbeat half that casts tech giants as benevolent, world-changing innovators.” 

In one chapter, Singer examines a 2019 Advanced Placement Computer Science course developed by Apple that teaches students how to code in Swift, Apple’s programming language for building iPhone apps. It’s as if schools embraced a “Raytheon-branded AP physics course or DuPont chemistry course or Pfizer AP Biology course,” she writes.

Singer finds both problematic and promising moves by tech over the past decade or more. While she calls out the nonprofit for skewing the idea of computer science to coding that often helps serve corporate interests, she lauds its founder, Iran-born entrepreneur , for making coding appealing and fun to millions of young people, telling ĂÛÌÒÓ°ÊÓ’s Greg Toppo that U.S. public schools likely wouldn’t offer computer science at this scale if not for Partovi.

“I don’t know anybody else that’s been able to do that,” she says. “I think that did change the world.”

Ahead of the book’s release on Tuesday, Toppo and Singer talked about her findings and the lessons that the tech industry is now taking into the fight to get into schools. 

The conversation has been edited for length and clarity.

The book spends a lot of time with Partovi and Code.org. Would you say it is basically a positive influence?

Natasha Singer: Code.org catalyzed computer science in schools. They also partnered with a lot of big tech companies and brought in commercial things — the Hour of Code has the lessons and Elsa the lessons and the Microsoft lessons and the Code Your Own lessons, and so it also brought in a kind of commercialism that I have questions about. I don’t think we would have computer science at this scale in schools if not for Code.org.

One of the key takeaways of this book is that there are other ways to do computer science in schools. In fact, if there’s a hero, it’s , who says Code.org and other groups are mostly just getting middle-class white boys interested in coding. Can you tell us about her? 

Jane Margolis is an incredible researcher. She grew up in Berkeley, Calif., during the ’60s, and she went to UC Berkeley. She was a Vietnam War protester and then she was a labor activist. She worked for the telephone company — at the time, it was Pacific Telephone and Telegraph, and there she saw these gender biases at work because she worked as an operator, the person who sat at the switchboard with all those wires plugging in calls. Male employees were allowed to be linemen, and they would go outside and climb these telephone poles and install phones, and she was trying to unionize the women to say that job should be open to everyone. Her view was that women need to have technical skills and equal opportunity to technical careers. Eventually, she goes to Harvard and gets a Ph.D in education, and she’s looking at gender disparities. Harvard has this government department, and it was inhospitable to women, and so she did her thesis about why boys were shouting down girls and pushing girls out of political science. 

She starts in computer science in the ’90s at Carnegie Mellon University. Carnegie Mellon, as you know, is at the forefront of computer science education and robotics and computer vision. They were one of the first to start a dedicated college of computer science, as opposed to a major. But a couple of years in, they realize that only 7% of the people majoring in computer science are girls, and they can’t figure out why. 

, who’s the associate dean at Carnegie Mellon, hires Margolis, and together they do a study looking at why it is that women are not going into computer science and are dropping out. And they find there are all these factors, one of which is that families are buying personal computers. She found by interviewing male and female students that the boys would be like, “Oh yeah, my dad or my uncle went to Radio Shack and brought home this TRS-80 computer and put it in my room, so I fooled around and learned to code, and then my neighbor came over — he’s a software engineer.” So by the time the boys show up at Carnegie Mellon, many of them have already taken computer science in high school. They know what they’re doing. And some girls have no experience, so they’re already feeling behind by Day One. 

It wasn’t just that girls were left out. It’s that when girls tried to take computer science in high school, often the guidance counselors would say to them, “Oh, you don’t really want to do this,” or “You’re not going to do well,” or “It’s going to hurt your grade-point average.” There were all kinds of socio-cultural things going on â€” sexism and a whole bunch of other stuff that dissuaded girls. 

And so they wrote this report at Carnegie Mellon, and they did all kinds of things to attract girls to computer science. There had been a requirement that if you wanted to major in computer science, you had to have taken it in high school. But of course, girls were dissuaded from taking it in high school. So one of the things that Carnegie Mellon did, they looked at other students who might have been good at math or sciences but didn’t necessarily take computer science, or who were leaders in their schools — perseverance qualities that make you good at computer science. Within a few years, the percentage of girls taking computer science shot up. 

And then Jane Margolis was like, “O.K., we now understand how to get more girls in computer science, but the percentage of African-American students, Latino students, lower-income students is even worse. She then becomes a researcher at UCLA. She works with Los Angeles public schools. They find similar problems of Black and brown and low-income students being dissuaded from taking computer science. So they come up with a course to make computer science more welcoming. 

Her whole idea is that if you do not grow up with a computer in your house, or you don’t know how to code, and you show up to computer science class and the first thing is computer programming, and the second thing is computer programming, and the tenth thing is computer programming, you may be turned off before you even know what it is.

So they created a class to explain to students who were first-timers: How does the internet work? How does a computer work? What is a motherboard? What is a central processing unit? And how are you using tech in your life? And do you know how an email gets to you? How are platforms — at the time it was more like and Facebook — shaping your life as a student? And then eventually, when they learn a whole bunch of stuff about how the technology works and why it’s interesting, then they start programming. 

And it’s not programming for programming’s sake. They made wearable computers — they had wristbands that had LED lights so they could make creative patterns. They could spell out their name, these wristbands, but they had to use computing to make it work. They had to program the designs. They made a lot of progress with this course, not just in Los Angeles but in Chicago. It was called Exploring Computer Science, and it began to grow slowly. 

Then the tech industry comes along and says, “We have this crisis: We need more programmers and software engineers, and every kid needs to learn to code.” And a lot of those courses were programming-centric, and I have to say some of them also included social aspects such as the ethical use of computing. But this notion that we live in a world surrounded by machines and it’s crucial for young people to understand how these things work in order to have agency and navigate them, that’s the idea that Jane Margolis was arguing for 20 years ago.

I want to ask about this story from 2015 that you tell about the course, which brought together the National Science Foundation, Code.org, Apple, and others. Can you briefly tell that story?

There was a leader at the National Science Foundation named . She was running a program called . And she was charged with broadening participation at the university level — that is, getting more women and African-American students and Latino students and students with disabilities and Native American students and others in computer science, because they’ve been historically marginalized. And she’s watching what Jane Margolis and others are doing in Los Angeles and around the country. And she realizes that the problem starts before kids get to college. 

So she gives millions of dollars to universities to develop different AP courses. The kind of course she envisioned to broaden participation in computer science didn’t exist at universities because universities took hardcore first-semester computer science that was designed to weed out people, and so she funded different universities, including UT Austin and the University of Washington and Trinity College, to develop more like Computer Science Zero, an online course that would look similar to what Jane Margolis was doing. It would teach you how the internet worked, it would teach you some data science, it would teach you some programming, it would teach you to think about the social impacts of technology. It would be a new AP course called Computer Science Principles, and you could take it as the intro. And then if you liked it you would go on and take AP Computer Science, which was much more rigorous and much more programming oriented.

So she convinces the College Board that they need to do this. And then, from Jan Cuny’s point of view, out of nowhere, The College Board announces that it’s partnered with Code.org to launch , a Code.org course, and then they announce that they have endorsed specific curriculum, including the Code.org one, but they later endorsed an AP Computer Science course from , which teaches kids to code in , Apple’s programming language for iPhone apps, and they also endorse a Microsoft course called . And Jan Cuny — NSF has funded a lot of this — is really stunned because nobody had told her beforehand that the College Board was going to . But also: Why weren’t they endorsing the ones that she funded? 

So she called The College Board and was like, “You’ve got to endorse all these courses then.” And by the way, the [AP] course was a smash success. Hundreds of thousands of kids took it. It was an introduction to computing that was broad for a lot of kids who didn’t know what it was. It did diversify who was taking computer science, although it’s still very skewed toward boys. 

But her point was that if you are trying to make a broad course that tries to get more kids of different backgrounds into computer science, Apple products skew to higher-resourced schools and higher-resourced kids. I read some of the Apple textbook, and there are lines like, “You can use this to make an app for your iPhone or even your Apple Watch.” This Apple course assumes that you have an Apple Watch. And Jan Cuny was like, “That’s not what we were doing when we were trying to broaden participation in computer science.”

I went back to the College Board, and they said, “When you’re broadening participation in computer science, you have to broaden it to everybody, and there are schools that do have MacBooks, and so maybe this course will work for them, whereas other courses won’t. It’s an array; you don’t have to use this one. You can use all of them or different ones.”

But for me, there are anomalies with having companies operate in schools, and the AP Apple curriculum and AP Microsoft curriculum show the unusual ways that tech companies influence schools that many other industries do not.

Fast-forward to 2026: Now we’re dealing with AI, and obviously tech is reading from a similar playbook. But it feels somehow different to me. I mean, we’ve had social media trials and , and these tech bros aren’t the white-hatted heroes we used to think they were. Does this moment seem different to you?

This moment is different, but also similar. We are in a moment in society where a lot of people have increased distrust of these huge tech companies, and we see a wave of parents pushing back against tech in schools. But we also see waves of people pushing back against , against , that there’s this feeling that tech has taken so much, and that people feel like they have lost control — that tech is being enacted on them.

And then when we talk about the social media trials and the big verdicts against Meta, both the one with dozens of states where Meta now has to pay $17 billion, but also the one earlier in the summer where New Mexico got in a judgment against Meta, those companies are no longer our darlings. And parents are worried about the effects of unfettered tech access and kids’ compulsive use of social media and phones, and so there’s a different climate. And yet there seems to be a disconnect in some ways with AI because you see that Microsoft and OpenAI and Anthropic and Google are all competing to get their AI tools into schools, and you see school districts across the country say “We’ve partnered with OpenAI.” “We are a ChatGPT Pioneer School,” or “We’ve partnered with Microsoft, and we’re going to use 10,000 licenses of CoPilot.” It’s amazing to me. We have had these cycles of tech in schools where there are all these promises about the latest tech, laptops, learning apps, massive open online courses, virtual reality, big data’s going to revolutionize schools and like democratize access to education for kids and get them these great career skills, and we keep having these cycles. My concern is that we don’t learn anything.

One of the reasons I wrote this book is because it feels like we have collective amnesia. Now we’re in the sixth or eighth cycle of this — and shouldn’t we be asking questions about the push for AI in schools, based on what we’ve learned about the push for all the other things that came before? 

I visited this school in Newark where two teachers had launched an AI literacy class because they were worried their high school seniors were going to graduate into a world with AI and not know anything about it. I went to the first class, and the teacher said, “We need to be honest with you. You’re the first kids we’re gonna do this class with. We don’t know how to teach this. We don’t know what you want to learn. And a year from now, this is gonna look completely different.” And there was this kind of honesty and humbleness: We, the teachers, are in it together with you, the students, and we’re gonna learn how AI works. 

They learned rigorous stuff, and they also learned to make stuff with AI. They weren’t just critical of AI, but they did it together. And if kids found something interesting, they went with that. And it’s a really hard thing for schools to know that we actually need to prepare kids about AI because they’re using it already, but in fact, we don’t know what to do yet, and this is going to rapidly change. That’s a really big challenge.

I was heartened by the chapters near the end where you go into classrooms. People are really embracing a new way to do this stuff.  

You know, I feel optimistic and upbeat about this new generation of teachers that want to equip students with broad critical thinking skills — and that involves both learning how the tech works and learning to use the tech and make stuff with AI, but also to understand the provenance of these AI tools and to understand that we have companies with more power than most countries, and we’ve never had that before. They are setting the rules we live by. So you have to understand how the tools are steering us.

New York City is expected to release its school policy for artificial intelligence on Wednesday, with a ban on student-facing AI tools for 2-K through eighth grade, according to four people briefed on the plans and Education Department documents obtained by Chalkbeat.

Under the new rules, the youngest children in the nation’s largest school system — those in its early childhood programs, elementary schools, and middle schools — would not have access to generative AI, including chatbots and AI tutors, sources said.

Get stories like this delivered straight to your inbox. Sign up for ĂÛÌÒÓ°ÊÓ Newsletter

The guidelines bar schools from giving individual devices to students in preschools through second grade, though classrooms will be allowed to have smart boards, sources said. For students in grades 3 to 5, the policy recommends limiting screen time on individual devices to 30 minutes and 45 minutes for middle school students.

High schoolers are expected to have restricted access to AI, but might be able to use it for literacy, career readiness programs, and in five pilot programs at an unspecified number of schools, according to the Daily News, . Screen time recommendations depend on subject matter and school approach.

The restrictions would represent a significantly more aggressive stance against AI and screen time in schools than the city had taken. Education leaders across the country are likely to be watching as they also grapple with how students should interact with the fast-changing technology.

“Younger students need hands-on learning, social interaction, play, and teacher-guided instruction,” Education Department officials wrote in a presentation outlining the policy obtained by Chalkbeat. “Routine individual screen use can crowd out those experiences.”

In high school, the presentation continued, “students need guided preparation for a world and economy increasingly driven by AI and new technologies -paired with instruction on how AI works, its biases and risks, and how to evaluate its outputs.”

Teachers will be allowed to use approved AI tools for such instructional tasks as lesson planning, translation, and drafting communications, according to the presentation. They will not be allowed to use the technology for grading, behavior monitoring, counseling, or developing special education plans.

City officials are in the process of creating a task force that will offer input on the policy’s implementation going forward, multiple people briefed on the Education Department’s plans said.

An Education Department spokesperson did not respond to requests for comment.

Multiple people said there would be some exceptions to the policy to account for regular assessments in reading and math that students typically complete on computers, as well as for some student groups, including English language learners and students with disabilities.

When the city unveiled its preliminary AI guidelines in March, forcing the city to rethink its approach and delay its timeline for a comprehensive policy by about two months, leaving little time for schools to prepare for the change before the first day on Sept. 10. It’s unclear how the Education Department plans to enforce the policy.

The initial guidance used a traffic-light framework to delineate risk levels of different AI uses in the classroom, mainly by teachers. Student AI use — one of the thorniest issues educators and families are currently confronting — got little attention.

The city seemed caught off guard by the momentum of grassroots parent groups calling for an AI moratorium, with thousands of people signing a petition to stop the use of generative AI in schools and scores voicing their concerns at school board meetings.

More than half of City Council members use in schools, citing concerns about learning and mental health. Samuels acknowledged in May that indicating that the final version would have more stringent rules on AI use, especially for younger students.

In July, as the city hashed out its new guidance, until officials released the new policy, who worried they would not have access to key software programs, such as electronic gradebooks and platforms to communicate with families, when welcoming students back to school next week.

Chalkbeat is a nonprofit news site covering educational change in public schools. This story was originally published by Chalkbeat. Sign up for their newsletters at

In concert, state bodies governing public pre-K through Ph.D. students are implementing policies addressing safety and classroom use of artificial intelligence.

Next week, the Florida Board of Governors, which controls Florida’s 12 state universities, will review language directing each campus to “establish policies requiring all course syllabi to include a disclosure of faculty use and permitted student use of artificial intelligence.”

The decision starts with the Task Force on Artificial Intelligence and Cybersecurity, which the Board of Governors created earlier this year to address the emerging technology’s permeation into higher education.

Policy proposals extend to Florida’s youngest students, too.

The Florida Board of Education, which oversees preK-12 and the Florida College System, has a vote scheduled to direct K-12 public and charter schools and colleges to regulate AI use.

The Board of Education will vote Sept. 16 during its meeting at Polk State College whether to require public school districts and charter school governing boards to amend their internet safety policy to include AI.

The rule is not expected to require legislative ratification and would have to be adopted by school boards by July 1, 2027. 

Districts and charter schools would, at a minimum, make policies to “protect students, parents, and instructional personnel from unlawful, deceptive, unsafe, or unauthorized uses of artificial intelligence systems, services, applications, and tools used in connection with educational programs and operations in Florida public schools,” according to a draft copy of the rule. 

The idea is to “ensure that any approved use of artificial intelligence protects education records and student information, while supporting student achievement, civic literacy, academic excellence, and workforce preparedness.”

More specifically, schools would notify parents when teachers approve an AI instructional tool for use, including the name of the platform and classes in which it would be used and the nature of student interaction.

Parental protection

Gov. Ron DeSantis’ administration did not return a request for comment.

The Florida Education Association supports guidelines for protecting children.

“Every parent and guardian wants their child to succeed in their public schools. In today’s world, technology is a key part of what and how our students learn, and, like with all new technology, there must be rules and guidelines that center on protecting our children,” FEA President Andrew Spar wrote in an emailed statement to the Phoenix.

The policy must require a process for parents to object to AI use and provide alternatives to the tool.

Schools would be required to conduct additional reviews for AI tools used in PreK-5 to ensure it is age appropriate. 

The measure includes language aiming to protect the teaching profession from being replaced by computers. It would “ensure that the use of AI instructional tools is limited to supplementing, and not replacing, the professional judgment of teachers and the primary right of parents to direct the upbringing and education of their children.”

Further, the policy must ensure AI is “accurate, transparent, viewpoint neutral, accessible to students and educators with disabilities.”

Schools must make publicly available approved AI instructional tools and policies. Also, the tools must maintain student interaction for at least 30 days in the interest of accessibility to parents and district administrators.

However, the data collected on students must not be used for profit or to profile students to train commercial AI models. 

The state is proposing, the regulation draft says, that schools not deploy AI to:

• Meet a student’s social or emotional needs, including Social Emotional Learning.
• Simulate friendship, companionship, or an emotional relationship with a student.
• Employ relationship-building or anthropomorphic design features for the purpose of encouraging a student to continue interacting with the system or tool.

The rule would impose training for teachers and administrators on the safety, risks, and compliance regarding AI. 

The policy must address plagiarism, permissible classroom use, verifying student work, and demonstrating mastery without use of AI.

“At the Florida Education Association, we believe that any policy the state of Florida enacts must ensure parents and guardians are advised of all potential uses of AI in a public school, that student data is protected, that the use of AI and new technologies is safe, that use of technology in the classroom is age and need appropriate, and that educators are protected and that additional work is not placed on them without additional support,” Spar said.

College system

The 28 state colleges have AI rules on the Board of Education docket, too. 

Those rules would place responsibility to create AI policies with college system boards of trustees

The Board of Education expects the 28 boards of trustees to create policies “that provide for the use and limitations of use of artificial intelligence tools in order to ensure quality of education for students, efficiency of operations, and a safe learning environment for the whole college community.”

As with the pre-K-12 policy, the college policy would distinguish between instructional AI tools and other technologies. 

AI instructional tools could not “meet a student’s social needs; simulate friendship, companionship, or an emotional relationship with a student; or employ relationship-building or anthropomorphic design features for the purpose of encouraging a student to continue interacting with the system.”

The policy calls on colleges to protect student privacy with AI, allow notification of parents of minors taking college classes whether their courses involve AI, and factor for academic integrity and dishonesty.

The policy calls, too, for building a framework to helps students understand what AI is and how it incorporates into their education and building AI skills.

This was originally published by Florida Phoenix.

is part of States Newsroom, a nonprofit news network supported by grants and a coalition of donors as a 501c(3) public charity. Florida Phoenix maintains editorial independence. Contact Editor Michael Moline for questions: info@floridaphoenix.com.

Wendy Graham’s two kids first encountered Amira, a popular artificial intelligence tutoring platform, in 2025, during a summer reading program in New Mexico. She also used it at home with her daughter, now a fourth grader.

But she found that the purple-haired avatar with the big, round glasses didn’t seem to understand where her daughter was struggling. When she stumbled over “usually,” the “learning agent” wanted her to repeat the word and spend time defining it. 

“She knew what the word meant; she just didn’t decode the word,” Graham said. Amira, in her view, “got hung up on all the wrong things.”

Her son, on the other hand, never gave Amira a chance.

“He just looked at me and looked at the screen like, ‘This isn’t happening,’ ” Graham said.

Her kids’ lack of enthusiasm for the program reflects what researchers have concluded so far about AI tutors. While some states are spending millions on Amira and other virtual tutors, most kids aren’t sticking with them long enough to benefit. For now, high-dosage tutoring models proven to be effective are human-led and depend on “a relationship that keeps students showing up and engaged,” according to from the National Student Support Accelerator, a Stanford University center that conducts tutoring research. 

“We don’t have solid research showing that AI tutoring can work in the U.S. at scale,” said Susanna Loeb, the center’s executive director. Researchers’ “biggest finding is that the take-up is very low.” 

Amira’s leaders, however, say their platform was based on extensive prior research showing AI tutoring can be effective . 

“No other scalable instructional intervention has been interrogated as many times, by as many independent teams, with this consistency of positive impact,” CEO Mark Angel said in a statement to ĂÛÌÒÓ°ÊÓ.

The brief comes as Amira, which records students’ voices to test their reading progress, faces increased scrutiny from districts in New Mexico. Parents expressed over how long the recordings are stored and what Amira does with the data. The New Mexico Public Education Department to use the program three times a year to assess K-2 students’ literacy skills. But in response to the opposition, it is allowing districts to , and Amria says it does not use the recordings for . 

Last month, the Las Cruces Public Schools, where Graham’s children attend and where she also teaches high school history, told parents they have disabled the recording feature and that they could request a paper assessment instead of Amira’s digital version. Officials also said they would discontinue use of Amira’s tutoring program.

“Teachers,” the memo said, “will continue to monitor progress and provide support for students to become proficient readers.”

‘Supercharge teachers’ efforts’

New Mexico parents’ reactions to Amira reflect broader reservations about the use of AI in schools. In from the Century Foundation, a left-leaning think tank, 84% of Americans said they’re worried about tech companies using students’ personal data to make a profit, and nearly half want AI use in the classroom kept to a minimum. 

The vast majority of survey respondents, 81%, expressed concern that teachers are being pressured to use AI even if it hasn’t been proven to help students learn. 

But Angel, with Amira, said the tool “can supercharge teachers’ efforts” by allowing them to listen to every child read aloud and identify struggling readers before third grade.

In 2024, the Duval County Public Schools in Florida signed a contract with Amira to help about 2,600 students in grades second through fourth reading below grade level and then tied half of the $100,000 contract to students’ progress. According to a of the work, more than 1,200 of the students met their goals for oral reading fluency, which exceeded the expectations of the contract.

Five hundred students made gains on state tests and about 300 reached the proficiency level, but the report doesn’t identify whether those results met the targets Amira was expected to reach.  

Graham, who previously taught fifth graders, appreciates what districts using Amira hope to accomplish.

“I think they’re trying to take things that have historically been done by teachers and streamline it to free up teachers’ time,” she said. But she still thinks states and districts aren’t getting their money’s worth. “Kids don’t do their best for robots; kids do their best for people.” 

Tutoring programs, however, can benefit from AI, the Stanford team concludes. A 2024 study, also from Stanford, showed that human tutors can become more effective with prompts from an AI tool. Tasks like scheduling, tracking attendance, analyzing notes from sessions and giving students additional practice outside of tutoring sessions are other ways that AI can support tutoring, according to the researchers.

The North Little Rock School District in Arkansas used , an AI platform that measures kids’ reading accuracy and fluency, as part of a research study. Students were expected to read to the AI program for 20 minutes, read a physical book independently for 20 minutes and then write about the topic, said Carmen Langston, the district’s literacy coordinator. 

On the screen, students would find digital books that teachers assigned or could create their own version, picking animals or scenery they prefer. 

“The kids like to put their names in there,” Langston said, and those who used it for the recommended number of minutes saw progress in phonics. But the district, she said, is trying to limit screen time. “We tell our teachers that if you are doing [an] intervention, it has to come from a human teacher.”

‘Engagement was thin’

The Stanford center’s research summary follows its of an AI tutoring program showing that students averaged about two minutes per week in one district and about five minutes in another — well under the two 30-minute sessions they were expected to complete each week.

A released last month tested the use of Khanmigo, Khan Academy’s AI tutor that relies on OpenAI’s technology. During the 2024-25 and 2025-26 school years, researchers Philip Oreopoulos and Nina Low randomly assigned students in 18 Hamilton County, Tennessee, middle schools to use the program for math practice. But while nearly all students with Khanmigo used it at least once, “engagement was thin,” they wrote. 

Students only consulted Khanmigo a third of the days they practiced math and just 17% of the times they made a mistake. Learning gains were about the same as they would have been had students not used the program, the researchers wrote. 

The results disappointed Sal Khan, who told that for students, the program’s release was a “non-event.”

Education Secretary Linda McMahon recently discussed the use of AI during a , saying there are not yet “a lot of metrics” to determine its value in the classroom.

“Are we seeing better outcomes in schools? And if we’re not, then they should be pulled out,” she said, adding that she thinks a parent who is dissatisfied with an ed tech tool “has every right to go in and say, ‘No, I want something different for my child.’ ”

The secretary, however, has repeatedly praised , a chain of for-profit AI-focused schools. Students spend two hours learning from AI and spend the rest of their days learning “life skills” like running a food truck or producing a musical, according to the website.

“The beauty of that is that the student almost has a one-on-one tutor,” she said. “The teacher can monitor to see if the child is a bit behind and repeat information for them. If the child is really excelling, then their programs get sped up a bit, so it’s a great tool.”

State contracts

Several states feel the same way about Amira and have spent millions to make it available to districts. New Mexico’s contract costs about $2.7 million a year. Iowa in 2024 to make licenses available to nearly 250 schools across the state and boosted the contract by to expand access to more schools.

The North Dakota Department of Public Instruction has a $2.6 million to implement Amira and points to showing that the more kids use it, the better their performance gets. But the data was self-reported by Amira and didn’t include a comparison group. The department declined to comment on research showing low student engagement. 

Utah doesn’t have a statewide contract with Amira and didn’t endorse it. An of interactive software programs showed that students considered at risk for falling behind in reading who used Amira outperformed a similar group of students who didn’t use the program. But of the 8,036 first through third graders deemed at risk, only 249 met the recommended usage time.

“Therefore, these results should be interpreted with sample size in mind,” the Utah evaluation said.

Amira also commissioned an analysis of the AI tutor’s use in Louisiana schools. showed better results for students who spent more time on the platform. That makes sense, Stanford’s Loeb said, but doesn’t prove the “causal effect of Amira.” The state spent in COVID relief funds on the program and in December, authorized another for additional AI tools, including Khanmigo and Writable.

Ted Beasley, a spokesman for the Louisiana Department of Education, said high-dosage tutoring has helped drive recent improvement in reading on the National Assessment of Educational Progress. 

“As we’ve expanded tutoring statewide, AI programs like Amira give school systems another option to complement that work,” he said. “A number of school systems have chosen to utilize this tool and have found it useful.”

As new research continues to emerge, other states are exercising caution. North Carolina’s recently for fiscal 2027 includes $500,000 for a dual enrollment math course using Khanmigo, as well as $2.5 million for Khan Academy partnerships with schools. Khanmigo can be turned off if districts decide. Lawmakers, however, originally recommended for the AI program.

Loeb still sees AI tutoring as promising, but said purchasing Amira or Khanmigo is not the same as “buying tutors for everybody.” She stressed that careful execution matters. That’s also what a on “intelligent tutoring systems,” one of the studies that inspired Amira’s founders, concluded. Positive results depend in part on “the adequacy of program implementations,” the researchers wrote.

“If you choose a tool that has some evidence or that you think could be effective, you have to figure out how to actually get kids to use it,” Loeb said. â€œIt’s going to take a teacher, a parent or some school person setting up the context where the students feel motivated to do the work.”

But for that promise to be fulfilled, students have to use the AI tools. New research suggests that may be much harder to achieve than schools assume. 

In a describing the results of randomized controlled trials led by the , 355 students in grades 1 to 5 in two school districts were assigned either to use an AI literacy tutor independently or to use the same platform with support from a human tutor. In both, the goal was to improve reading achievement.

The human tutors encouraged students to participate, helped troubleshoot technology problems and established group norms, while building rapport and accountability. Their sessions followed a consistent structure: a brief check-in, time for students to work on the platform and final reflection. In the group with no human support, students were expected to work on the AI platform on their own. Each group was scheduled to use the tool twice a week for 30 minutes per session.  

The clearest finding from the study was that all students, regardless of whether they received human support, used the AI tutor far less than intended. For example, almost half the children in the independent-use group did not use the AI tutor at all. Those who did use the platform spent just two to five minutes per week, on average.

Students using the AI tutor with human support spent only slightly more time using the tool, but the effects were small: an increase of just one minute a week in one of the districts and 4.4 more in the other. 

There are a range of potential reasons for this low engagement, from students not finding the tool compelling to preferring to ask a teacher for help or needing more encouragement and accountability.

For both groups of students,  usage remained far below the 30-minute-per-week dosage recommended to reap academic benefits. At these levels, students likely did not receive enough of the intervention for gains in reading achievement to emerge.

These findings underscore two important points: Access to AI tutoring does not ensure meaningful use, and, while human support can increase student engagement, it may not be enough to generate measurable learning gains. The study highlights the real-world implementation challenge inherent in student-directed AI tools: Kids need to be motivated to engage and keep coming back, or these products will never deliver on their promise. Human support can help, but in this study it wasn’t enough to bring usage close to recommended levels.  

But there are other promising approaches that flip the model: rather than asking AI to replace human tutors, AI can help them become more effective. In , an AI tool that provided real-time instructional suggestions to human tutors improved student mastery, with the largest gains occurring among less experienced tutors. This suggests a for the future of personalized learning, one where AI and human support work together to meet the needs of tutors and individual students. The right mix will vary depending on the subject and grade level. Assuming stronger engagement, some students may thrive with primarily AI-supported learning. Others may benefit from AI tutoring paired with periodic human encouragement and accountability. Still others will require sustained relationships with human tutors, teachers or mentors. 

That is where the field should focus next.

But before schools and districts implement AI tutoring tools with the expectation that they will transform learning, they need more, better evidence about what it actually takes to engage students. To date, the research remains surprisingly limited. A by the SCALE Initiative identified more than 800 studies examining AI in education, yet only 20 used rigorous designs that allowed researchers to examine the causal effect of the AI tool on teaching and learning. In other words, enthusiasm for AI in education has far outpaced rigorous evidence about whether students will use these tools, and whether they improve student learning. 

Widening the evidence base is essential. Funders, researchers and education leaders should prioritize uncovering which kinds of human supports are needed for which students, under which conditions and whether they generate enough meaningful engagement to improve learning. Also important is research to inform decisions about which models are both effective and scalable. Organizations like and the , both Overdeck Family Foundation grantees, are beginning to explore more cost-effective, hybrid approaches that use AI to personalize instruction while strategically deploying human support where it matters most. Whether these hybrid approaches can improve learning at scale remains an open question, and an important one to test. 

Education has seen this story before. New technologies arrive with the promise of reaching more students at lower cost, only to produce disappointing results because of implementation challenges. And while AI has enormous potential to expand access to personalized learning, there is no reason to assume it will succeed simply because the technology is more advanced.

The next generation of research should help education leaders and practitioners understand not just how to build better digital tools, but how to build learning environments where students actually use them and benefit.

Disclosure: The Overdeck Family Foundation provides financial support to ĂÛÌÒÓ°ÊÓ.

Just as the technology has rewired the relationship between young writers-in-training and their teachers, it has also changed how student journalists and their editors work together — or, in some cases, don’t. 

And AI has created headaches for these journalists that mirror those afflicting their professional peers, with school newsrooms struggling to figure out how to respond.

As with their schools more broadly, student newsrooms have spent these years deploying surveillance software and, in many cases, drafting zero-tolerance policies for AI. Different newsrooms have developed different answers to the same basic question: How much do we trust teenagers to not let a chatbot do their work? 

ĂÛÌÒÓ°ÊÓ spoke recently to three student journalists from three very different newsrooms about their experiences:

• Cid Frydman, a rising senior at Berkeley High School in California, who writes for the ;
• Veronica Mollod, a rising senior and current events editor at , the student newspaper at the High School of American Studies at Lehman College in the Bronx, N.Y.;
• Tova Weiss, a recent high school graduate in Ann Arbor, Mich., who has split her time between Pioneer High School and . Weiss has written and edited for the and , a publication written by and for Jewish teens.

Here are eight-and-a-half ways — including one counterintuitive, not-for-attribution observation — in which AI is changing their realities:

1. In a few schools, AI has overturned the culture of newspapers.

Frydman, the Berkeley student, said strict AI use policies at her publication have bred resentment among student writers. 

“I think a lot of the people higher up on my paper, and in journalism in general, don’t have a lot of trust in writers anymore, especially with so many AI resources that are out there,” she said. 

Editors are “very harsh on writers” over worries that they’ll use AI, keeping strict controls over documents, she said. They’ll poke around writers’ search histories and analyze their typing speed and word choices, matching drafts to previous work. 

“It’s very strict when you put it all together,” Frydman said. “They don’t have a lot of faith in our ability to actually produce real journalism, and so just assume the worst automatically, which can be a little frustrating when you’re just trying to do your best as a writer.”

At the Berkeley High newspaper, staffers have been let go over suspected AI use. 

Frydman sees the high cost of such heavy-handed policies: “We’re here because we love journalism,” she said. “We’re here to write about things that we’re hopefully passionate about, and the heavy approach of ‘There’s no tolerance whatsoever,’ and not even giving writers the grace and the trust to make their own decisions and do this work on their own — just minding it and monitoring it so heavily, I think it’s very discouraging.”

More broadly, getting caught using AI to write academic assignments carries weighty consequences, she said. You’re unable to get a letter of recommendation from any teacher, Frydman said. 

And a single infraction travels fast: Get caught using AI once, she said, and teachers will tell all of their colleagues “to watch out because this student has used AI before — they might use it in your class.”

2. AI is changing how — and how much — students write.

Frydman said she’s adopting what’s basically a minimalist approach to writing these days, holding back details so she doesn’t draw suspicion from editors or teachers. 

Though she doesn’t use AI for writing, she said, “A lot of times, I find myself sort of slimming down my writing to not go so into detail on things because it’ll be suspected to be something that’s not original. You don’t have a lot of room to be creative in this process, so it’s definitely been a little challenging.”

Ann Arbor’s Weiss said she’s also seeing many young journalists “adjust their writing voice” to avoid being flagged as relying on AI. Many students, she said, now intentionally add typos to their writing to support its legitimacy. “It’s absolutely devastating to watch how many people are intentionally damaging or diminishing the quality and authenticity of their work,” she said. 

At Berkeley High, school-issued Chromebooks are locked down to block access to ChatGPT, Claude and other major AI tools outright — and any attempt to reach them gets flagged. “It notifies the teacher through this program that we have called Go Guardian, which is embedded into a lot of the Chromebooks, where they can sort of see what you’re doing and be able to call you out on it,” Frydman said.

3. A few newsrooms are writing AI policies.

At Youth Environmental Press Team, where Weiss is an editor, staffers have been talking “in pretty much every director’s meeting” about the implications and problems surrounding AI. “We are an environmental publication, which means a lot of us are very climate-conscious.”

The environmental impacts of AI data centers are a cause of great concern, Weiss said.

After fielding a wave of questions from reporters, the publication developed formal guidelines rather than instituting an outright ban on AI. “It was always very clear for us that we would not publish or even entertain the idea of publishing anything that was written by a chatbot or any kind of generative AI,” Weiss said. But they soon got queries from reporters asking if they could use AI for outlining, for formatting statistics, for interview transcriptions and the like.

The policy that emerged prohibits AI-written and AI-edited work, including work that heavily uses tools like Grammarly. But transcription services and basic spell-check are still allowed. 

The organization, she said, wants to make writing “as accessible as possible so people can have their work published — but not if that means it’s not really their work, and especially not if that’s at the cost of the environment.” 

4. Other newsrooms rely on experience and vibes.

Weiss helped craft the formal AI policy at the environmental journal. At jGirls+ Magazine, by contrast, she said there’s no written policy. But in practice the standard is clear: Submissions and even staff applications that appear to be AI-generated are not accepted. In one case, she recalled, an obviously AI-written editorial board application — with a stiff style that lacked personal voice — tested with an AI detector as 100% AI, “which was really something none of us had expected.”

Weiss said AI detectors should be used only “when there is already prior reason to suspect something might have been AI generated” — a student with a track record of misusing AI, for instance, or because something just simply reads a little “off.”

5. In most cases, AI detectors are widely seen as unreliable.

Weiss described a tiered system in the environmental outlet’s workflow, with writers operating on an “honor system” but with student editors flagging “sketchy” pieces such as those that are missing statistical sources or those that quote dubious ones — or writing that doesn’t sound like a typical teenager’s.

“The thing about student voices is typically each writer has a distinctive style of writing,” she said. “We’re high schoolers: We have typos, we make mistakes in there. And if we’re finding something that’s void of personal style or just generally reads in a way that doesn’t seem like it was written by a student, we’re going to look into that further.” 

But she and others worry that using an AI detector could unfairly flag strong writing as suspicious simply because it’s polished — a phenomenon that happens more than you’d think in many schools, Weiss suggested.

AI detectors aren’t really reliable, said Frydman, the Berkeley senior, “so even if you write something really well, it’ll just automatically assume that it might be AI and your writing will get flagged, which has happened to a lot of my friends.” 

6. A few simple rules can keep AI from taking over, journalists say.

Mollod, the New York City journalist, said a somewhat arcane editing rule at her publication basically serves to help keep AI-generated content at bay: At Common Sense, every article is required to include a quote from a student in each grade level — a structural feature that makes it hard for AI to intrude, since students typically remember what they’ve said to a reporter. 

7. In a few cases, teachers are using AI more than students.

The journalists described several instances in which teachers, not students, were the ones turning to AI, relying on the technology to formulate lesson plans, create assignments and grade them.

One student who asked not to be quoted directly spoke of a teacher who “will just copy and paste straight off of Chat GPT or Claude, and just give us assignments that are so clearly AI that it’s ridiculous.”

Others rely on it for grading assignments, which “leads to a lot of people getting unfair grades, because AI can mess up,” said the student.

The more teachers rely on the technology, “the easier it is to just forget that you’re there for a reason, you’re trying to do your job for a reason — and AI should not be teaching kids Spanish or English or math or anything like that. It can’t reproduce real thinking. So it’s definitely a little dystopian.”

8. For many students, refusing AI isn’t a school rule. It’s personal.

Weiss, the Ann Arbor writer, said she has never used Open AI’s popular ChatGPT, due to environmental concerns over data centers as well as personal concerns over AI more broadly. “It just does not align with my morals at all,” she said. 

But casual AI use has become normalized among many of her peers, she said, with students openly saying, sometimes within earshot of teachers, that they ran an assignment through a chatbot the night before it was due.

The best way to fight AI, she said, is basically to boycott it. “The strongest way anyone can combat artificial intelligence is just by refusing to have anything to do with it,” she said.

While AI-generated writing more broadly can damage the way that people learn and is worth thinking about, Mollod said, she’s not sure that it has gotten to the point where “just everyone is just cheating, and no one’s really actually taking their work seriously.”

She reads the coverage and believes most people who don’t spend time in schools overestimate how much teens use AI. “People not in school are like, ‘Well, I didn’t have this great tool. If I had it, I wouldn’t have done anything,’” she said. But relying on the technology, which showed up in the middle of her eighth-grade year, isn’t yet common practice for most students she knows. 

Whether future students will warm to it more isn’t yet clear, Mollod said.

Weiss, the Michigan student, said her work — especially at the environmental publication — has taken on a new significance in the age of AI, pushing to celebrate and highlight “authentic youth journalism and youth narrative” in the face of encroaching technology. “In my view,” she said, “demonstrating our talents as humans and as youth is a form of silent protest against AI by proving that artificial intelligence isn’t necessary to produce professional-quality journalism.”

Bonus: Non-AI cheating abides. 

On background, one of the students said she thought a lot of the news coverage about AI cheating, while serious, simply misses the point. The old-fashioned kind of cheating, for now, remains much more common. In 2026, students are still relying on copying each other’s homework, just as generations before them did. 

“Mostly it’s just, like, circles of people, where there’ll be a math class and five people do the homework — and the other 20 got it from this person, [who] got it from this person, [who] got it from this person — which has always kind of been going on.”

On the second day, one of the teens was fed up. Gemini was broken, she told me. It kept asking her questions instead of answering them. She was adamant that ChatGPT was a better tool, and that we should use it instead. 

What she didn’t know was that I had told Gemini to behave that way, to question her about her daily reading instead of just handing her answers. Then she scrolled up and read the instructions I had written into the prompt, and her annoyance turned to understanding. She wasn’t fighting a broken tool; she was being forced to pause and think.

The next day, she scored 100% on a quiz covering the prior day’s reading. She started leaning into our daily lessons, taking the time to answer questions thoughtfully instead of rushing through them. During our final session, she was one of the last students to turn in her quiz, making her mom wait a few extra minutes while she finished. Unheard of, especially over summer break.

That moment showed what’s possible when AI tools are built to make students actively think rather than think for them.

Over the course of a week, I gave the students a series of short readings about how their own brains work, and each day they used AI in a different way to deepen their understanding. Some days, I had the tool summarize the reading and told them to simply use that summary to build their understanding of the material. Other days, the tool had them engage in a Socratic Q&A. And some days it made them write their own question first, letting their own curiosity lead the process.

Then I quizzed them on what they remembered, both after their study session that same day and again a day later.

The results were fascinating. On the same-day quiz, students performed about the same, regardless of whether the AI tool had summarized the reading or questioned them. But a day later, the students who had engaged in the Socratic Q&A remembered more, scoring, on average, 83% on a quiz about the material they had reasoned through, versus 67% on the material the AI had summarized for them. More than two-thirds of the students said the version of AI that questioned them helped them learn more than the one that just summarized their readings.

Throughout the week, the students’ perception of their own skill in using AI as a study partner rose sharply. But when I lined those ratings up against what they could actually remember, the two had almost nothing to do with each other. How much students’ confidence in using AI grew had almost nothing to do with how much they had retained. Feeling like you have learned is not the same as actually learning.

This should worry anyone who cares about education. AI is everywhere. Students carry these tools in their pockets. School districts are slated to spend in the coming years, without a clear idea of what they should be evaluating those contracts on. My week-long pilot program taught me that AI tools cannot be judged by student engagement or satisfaction alone. Success cannot be gauged by whether students finish a problem set, or whether they enjoy a particular lesson. It must be measured by what a student retains after the tool disappears.

My students told me one more thing worth highlighting. They want to learn to use AI as a thinking partner, but no one has taught them how. When they play around with AI to get answers, they often can’t obtain explanations at their own grade and language level. They told me it is sometimes harder to get what they need from these tools than to study on their own. The students who could benefit most from personalized learning get the least usable output, perpetuating the very inequity these tools are supposed to close. 

That is why AI literacy must be taught in schools. Kids with tech-savvy parents are already learning to use AI well; the rest are typing in prompts and hoping for the best. School is the one place that can close that gap.

A College Board last fall found that the share of high schoolers using it for schoolwork had climbed to 84%, with about half using it to brainstorm, revise essays or research topics. A follow-up found that high-achieving students — and those who are perhaps experiencing “senioritis” — are the most expansive users of AI. 

A new survey by the nonprofit Common Sense Media offers another set of key data points, looking at teens’ own sense of how they use the technology — and what AI is doing to their thinking.

The nationally representative sample of 1,017 U.S. teens ages 13 to 17, conducted in April and May, found that 70% now use AI for schoolwork, but that fewer than one in three say a teacher has ever talked with them about using it safely. And just one in four have discussed with a teacher how to tell whether AI-generated information is accurate.

Here are five things to know about the findings:

1. Most teens say they’re using AI to support their thinking, but nearly as many see it as an answer machine.

Among the 70% of teens who use AI, 77% say they use it as a support tool, using it for brainstorming, checking their work and getting feedback on writing, for instance. And 66% say it helps them understand schoolwork, not just finish it faster.

But 63% separately report going straight to a chatbot for an answer rather than working their way to it themselves. Of those, 25% use AI output as-is, while 31% rewrite it to sound like their own voice. Another 35% “improve it with what they know.” 

And yes, the 77% and 63% figures add up to more than 100%. That’s because many teens say they are, at different times, different kinds of AI users, either looking for support and brainstorming or simply zeroing in on a quick answer, depending on the assignment. 

Former U.S. Education Secretary John King, an advisory board member for the nonprofit’s , said the 63% figure is worrying “because we can’t allow AI to become a substitute for students doing the learning themselves.”

Much of that will fall to teachers, he said, who must change their basic practices. “Some of that is old-fashioned blue books and saying, ‘We’re going to do some of the writing in class,’ complementing a written assignment with an oral presentation in class so you know that students are really doing the work.”

He acknowledged that changing course design “is really hard,” there’s really no alternative. “I think this is really a moment where the entire K 12 sector has to think through: ‘How do we help students understand AI safety and appropriate use of AI tools, and how do we adapt teaching and learning assessment so that students are doing the intellectual work?’” 

2. Teens know that AI takes a cognitive toll.

Many of those who rely on AI for schoolwork acknowledge that it’s changing how they think: 38% say having AI available leads them to develop fewer original ideas, and 39% say that when they use AI to complete assignments, they feel they’re missing out on learning.

Darren Lam, 18, a recent graduate of New York’s , said AI use was widespread among his classmates. “A very large percentage of people that I know either regularly use, or have at some point used, AI to just completely do an assignment for them,” he said. “So I think that the concern of AI being used as a cheating tool is definitely not invalid.”

But he said the bigger problem is students not coming clean about their motivation — or lack of motivation — to learn. “If students wanted to genuinely learn and were willing to be honest about their own work, then giving them free access to AI would not encourage them to cheat in the first place,” he said.

As it is, for many students school simply isn’t relevant, he said. “They think that many of the things that they learn in the modern education system aren’t particularly beneficial once they leave the four walls, and I think because of that they want to just skip through it as fast as they can. And the fastest way is by just getting AI to do all your work and then submitting it.”

3. School AI bans don’t really work. 

Among teens who use AI for schoolwork, 44% say they’ve had an AI tool blocked on a school device or network. Yet 59% of these students say they simply switched to a personal device with less oversight — and less support — than a school-managed device or network.

Lam, who plans to major in electrical engineering in college, said his Staten Island high school has a fairly open policy about AI — it’s full of future coders and computer engineers, after all — but he worries more broadly that schools limiting access to AI tools could backfire.

“Although it can sometimes prevent some harmful behavior that may arise,” he said, “it also prevents students from learning how to use it effectively, which is going to be a very pivotal skill coming into this job market.” 

4. Teens are clear on which skills matter most going forward.

Asked to name the skills that matter most for their future, teens zeroed in on a few key ones, dubbed “the human skills that AI can’t or shouldn’t replace” by Common Sense:

• 55% chose “working well with others” 
• 50% chose “fact-checking and critical thinking”
• 48% chose “reading and understanding complex information”

King, the former Education Secretary, who the State University of New York, said helping students develop these skills in response to AI will be difficult, but educators can’t bury their heads in the sand.

“I think a lot of teachers already, understandably, feel overwhelmed by the core challenge of getting students the academic knowledge and skills they need,” he said. Noting that many schools are still doing “recovery work” from the disruption of the COVID pandemic, he said, “I’m sure it feels like a lot. That’s certainly true for our faculty at SUNY, but I think the reality is these tools are here. Students are using them, and we have to respond thoughtfully.” 

5. AI literacy instruction in schools is the exception, not the rule. 

The new findings arrive Tuesday alongside the launch of a new Common Sense and a new AI and Student Well-Being professional development course. The curriculum offers a research-backed framework for building students’ AI literacy skills through two courses, one designed for K-8 students and a second for high school students.  

According to the survey findings, only 30% of teens overall say a teacher has ever talked with them about how to use AI safely — and just 26% say teachers have discussed how to evaluate whether AI-generated information is accurate or trustworthy.

Alex Kotran, CEO of the , said that’s especially troubling, given AI’s prominence in kids’ lives. “The fact that 70% of kids haven’t had a conversation about AI safety is a pretty big alarm bell, because it means the kids are using this stuff, and a lot of them are flying blind.”

Key to addressing the crisis of students using AI to offload cognitive effort, he said, is  training teachers how it works. “There is no solution that doesn’t involve a teacher just understanding how AI can be used, and through that understanding [and] adapting assignments and lessons.”

Like other observers, Kotran suggests that teachers use this moment to rethink assessments, turning to in-class demonstrations, talks and the like to ensure that students understand material. 

“Even if a student is going home and using AI, it’s not going to help them because ultimately they’re going to have to go and demonstrate their knowledge in the classroom,” Kotran said. “I think it is as simple as that. There are other ways to adapt assignments, but I think it’s really hard to adapt your assignments if you don’t understand what AI is actually capable of.”

It’s move-in week for college students across North Carolina. And once students have gotten settled into their dorms, their attention will turn to getting the classes needed to fulfill their chosen major. But which career paths are a safe bet in 2026?

U.S. employers in July. Job creation for May and June was also revised downwards, showing a combined loss of 103,000 positions.

A report by outplacement and executive coaching firm Challenger, Gray & Christmas found artificial intelligence remains , accounting for almost 11,000 of the announced cuts last month. It’s the fifth straight month AI has been the leading explanation that employers have provided for eliminating jobs.

“AI is shifting the labor market, it is not dismantling it,” Andy Challenger, chief revenue officer for Challenger, Gray & Christmas, in their latest report.

Technology, financial firms, and the government sector accounted for the most job losses in July.

Gerald Cohen, chief economist with the Kenan Institute of Private Enterprise, calls the latest employment data worrying.

Cohen said economists are actively debating whether AI is going to create massive job losses in certain fields – careers one would want to avoid – or if it will create new career pathways.

Careers to see major disruptions in the near future include marketing and public relations, customer service, technical writing, journalism and web development.

Cohen says ultimately, it may come down to how a person uses artificial intelligence to their advantage.

“If you’re a critical thinker, learn how to use AI and learn how to leverage it, [choose] whatever field that’s going to inspire you,” Cohen said in a call with reporters.

North Carolina State University economist Dr. Mike Walden said two or three years ago, he would absolutely have recommended a young person heading to college pursue a degree in technology. It’s a less safe bet these days.

“Now that’s up in the air because of AI,” said Walden. “AI is a technology that can replace people.”

Walden believes that while it’s unclear to what extent that may happen, AI is more likely to change certain jobs than to eliminate them.

“There’s a part of decision making that only humans can do,” Walden said.

Walden told NC Newsline the uncertainty AI creates may help promote social sciences.

“Social sciences are fields where you rely on human interaction, being able to assess other people, being able to work with other people,” explained Walden. “That’s actually something that I would tell students going to college, and say, certainly look at technology, but don’t forget the social sciences. I think that may very well have a rebirth in the new economy that has AI.”

Walden said skilled trades are another area where AI can’t easily replace humans.

“I’m also a big promoter of skilled fields that maybe require just community college,” said Walden. “We need entrepreneurs, we need plumbers, we need electricians. That’s not what a lot of people think about when they go to college, but prospects for those jobs, I think, are very, very good.”

For those newly graduated or early career professionals who are still struggling to land a job, Cohen says it may not be AI’s fault.

“One of the reasons why the unemployment rate for folks with bachelor’s has been rising faster than the unemployment rate with folks without a bachelor’s degree is not because of AI, but because of the impact of remote work and the fact that it’s harder to gain [certain] skills from, and it’s harder to learn from other workers when you’re remote,” said Cohen.

Cohen said he’s a big proponent of flexibility, but the truth is many employers are still grappling with managing and growing employees who aren’t on-site as the workplace continues to change.

This was originally published by NC Newsline.

is part of States Newsroom, a nonprofit news network supported by grants and a coalition of donors as a 501c(3) public charity. NC Newsline maintains editorial independence. Contact Editor Laura Leslie for questions: info@ncnewsline.com.

The New York-based company is developing AI tools to give teachers and students immediate feedback on papers, exams, recorded presentations and even artwork to the 400 public and private schools in the consortium.

Legend.org co-founder Matt Sornson said the company’s platform lets teachers upload student work — either digital files or even photos of handwritten work. AI will then use state, school, or teacher standards to evaluate the work.

The technology will go even further for mastery schools, which gauge how students improve toward mastering soft skills such as communication, collaboration and critical thinking. The AI tools will help teachers track progress on these more subjective skills using indicators educators and researchers are still developing.   

Sornson said that AI tools he is working on will evaluate communication skills, for example, from recorded verbal presentations.

“There’s some pretty pretty cool models around where you can do tone of voice, you can do body language and posture, you can do eye contact,” he said. “There’s a lot of really really cool things you can do with video.”

Sornson said AI would not judge whether a piece of art work is “good,” just technical aspects of it. 

For visual art, he said that could include composition, technical execution, use of materials, visual communication, and originality.

For music, Sornson said, AI could judge “rhythm, pitch and intonation
.” The technology could also “review recordings, video, written scores
and student reflections,” he said. 

He stressed that AI won’t be giving student work the final ratings, just acting as an aide to teachers.

“It does the first pass of assessing (their) rubric and what (they’re) trying to grade,” Sornson said. “(AI) would ingest those assignments, those artifacts, and then do the first pass of assessment. It’s always the teacher choosing the final grade.”

“We are just taking the first version, showing them where they should focus their efforts, and trying to save teachers time,” he added.

That middle ground could be an important one, as educators and researchers nationally sort out how AI can be best used to help teachers grade student work and offer meaningful followup instruction. Some researchers have found AI feedback too lenient to students or to

Several colleges and schools are trying to so it can help teachers and also protect student interests. 

Experts have stressed the importance of being transparent with students about how much teachers use AI and of making sure the technology isn’t showing any troublesome biases. Having humans make important decisions also matters, wrote Amanda Hirsch, an assistant director of the Center for Innovative Teaching and Learning at Northern Illinois in a

“As we stand at this crossroads of education and emerging technology, the question isn’t whether to use AI in grading, but how to use it ethically and intelligently,” Hirsch wrote.

The Mastery Transcript Consortium is a group of schools that has developed a new report card and transcript model that skips A-F grades in favor of tracking student progress on dozens of academic and soft skills, such as collaboration, communication and critical thinking.

In 2024, it merged with the Educational Testing Service, the company known for running the SAT and GRE college admissions tests, joining the Skills for the Future effort by ETS and the Carnegie Foundation for the Advancement of Teaching to better assess how well students are learning skills and to prove those skills to colleges and employers.

But ETS decided to focus on creating new assessments and less on report cards. It sold the consortium and its transcript model last month to the for-profit for an undisclosed amount.

Consortium founder Scott Looney, head of the private Hawken School in Cleveland, Ohio, said he is also excited about how AI can help teachers. AI, he said, can reduce the time teachers spend reading and grading student work and can offer immediate feedback to both students and teachers that they can discuss and shift quickly toward improving. 

Gauging how well students are doing on soft skills — such as self-direction, generating solutions or synthesizing information — isn’t simple, either, he said. Students build portfolios of work across multiple courses that show their progress on multiple skills. AI, he said, can evaluate progress across many pieces of work and summarize them much faster than teachers — and later college admissions staff — can.

“If we can make this easier for students to pull their portfolios together, easier for teachers to get feedback to the kids about the quality of the work in their portfolios, and then easier for the receivers of our transcripts that are full of portfolios to interpret them, I think this is going to go exponential,” Looney said.

Over time, Looney said, he hopes students could even submit work to the platform on their own for suggestions before turning it in to a teacher.

A key feature of the upcoming tools, Sornson said, is that student work can be measured against any standards a school chooses. That could include public breakdowns of skills and subskills — such as those from ETS and Carnegie, , or — or any that schools develop themselves.

Looney said colleges or employers might eventually be able to submit their own standards for skills they want from applicants for AI to use to measure student work from all four years of high school.

Though Looney said one of his campuses that uses the mastery report cards will try the AI tools out as Legend offers them, other schools in the consortium are still learning about the sale and haven’t said yet what they plan to do. 

“I would be surprised if schools decide to forsake the AI tools as they will be highly accurate and save students and teachers time,” Looney said.

Looney said he was concerned at first that the transcript model he created with other schools would turn into a for-profit product. But he said the partnership relied on grants to operate that are not sustainable. The computerized system that creates the transcripts also needed improvements Legend is already doing, making it easier for new schools to join.

The split from ETS and Carnegie is friendly. Both Looney and Laura Slover, who heads the Skills for the Future effort of ETS and Carnegie, will be on an advisory board to the new company.

My job is to tell a student, kindly, when an argument is not working yet. But the machine had just told them it was. Not because the machine was broken, but because it had been optimized to agree.

For two years, educators have warned students that artificial intelligence makes things up. It can give a wrong answer wearing a confident face, a phenomenon known as hallucination. But there is also a quieter failure we have barely named, and it is more corrosive because it feels like support.

Researchers call it sycophancy: when an AI system flatters, agrees with or validates a user even when a correction would be more truthful. A hallucination is a wrong answer. Sycophancy is a wrong relationship.

When students head back to school at the end of the summer, AI will already be part of their schoolwork. Last year, the ⁠ that 84%of high school students used generative AI for schoolwork, and a February ⁠ found that 54% of U.S. teens had used chatbots for assignments: writing, summarizing articles or videos, studying, and creating or editing images.

A from Turkey illustrates why this is a problem. Nearly 1,000 high school students were given help from an AI math tutor powered by GPT-4, the technology behind ChatGPT, while practicing math. They solved more practice problems with the tool. But when the tool was taken away for the real test, students who had used the less restricted version, which could provide more direct answers, scored 17% worse than students who had no AI help. In the same study⁠, a guarded tutor – designed to give hints instead of answers – avoided that drop in test performance.

AI can make learning look easier while quietly removing the struggle that makes learning stick. In writing and analysis, where judgment matters, it can even encourage students to do inferior work.

Learning runs on friction: the red pen, the Socratic question, the “not yet, try again.” Education often tells students they are not right yet so they can become right. A tool engineered to affirm removes a core foundation that education depends on.

In April 2025, OpenAI rolled back a version of GPT-4o after it had become, in the company’s words, “⁠.” The company later admitted it hadn’t tested for sycophancy before rolling it out to hundreds of millions of users. OpenAI pushed a temporary system-prompt fix, then returned users to the earlier version. The rollback fixed that release, but it didn’t address the broader problem of AI systems becoming too agreeable when users need correction.

The pattern is not limited to one chatbot. In March 2026, a Stanford-led team published a ⁠ across eleven leading models. When users asked for guidance about personal situations and relationships the models affirmed their actions 49% more often than humans did. In prompts involving deception, illegality or other harms – such as lying, manipulation or socially irresponsible behavior – they endorsed the behavior 47% of the time. People who interacted with sycophantic AI became more convinced they were right and more likely to trust and reuse the tool. 

A student who asks whether a weak thesis is strong, whether thin evidence is enough or whether a draft is ready to submit may not be looking for praise. They may be looking for a teacher. If the tool offers confidence instead of correction, the student never feels the discomfort that turns a half-formed idea into a real one.

I am not arguing that students should avoid these tools. But every student using AI should ask: When has this tool stopped helping me think and started flattering me into not thinking?

That confusion may cost the most among the students I work hardest to reach. At a historically Black university, I teach young people who are brilliant and underestimated, who are wary of how institutions judge them and who sometimes turn to AI because it does not seem to judge at all. 

For students surrounded by regular, trusted feedback, an over-flattering tutor may be an annoyance. But for those with fewer chances to get honest, trusted feedback, it can become a closed loop: The student asks, the machine praises, and no one tells the truth. 

A 2026 Stanford study of ⁠ makes that worry more concrete: identical essays received different feedback depending on how the student was described, with essays attributed to Black students receiving more praise and less substantive criticism. That is not reassuring.

So what do we do?

First, educators should teach AI as something to interrogate, not an answer machine. Assign students to make the model disagree with them, hunt for the moment it praised a weak argument, and grade that catch. “Where did it just agree with you?” is a teachable skill.

Second, instructors and advisers should change the question they ask about AI help. Not just, “Did it give you an answer?” but, “Did it make you explain your thinking?” “Did it ask what evidence you had?” “Did it push back?” “Did it make the work harder in the right way?”

That kind of friction has worked before when it is built into the tool, giving hints instead of answers, asking questions that make students show their reasoning and offering prompts that slow them down before they accept AI advice. In the Turkey study⁠, the guarded version that guided students instead of simply spitting out answers largely avoided the learning harm. Other studies have found that prompts requiring users to ⁠, ⁠, and make their own judgment can reduce blind acceptance of AI advice. Until every tool is designed to slow students down in those ways, adults must teach students to add the friction themselves.

We spend our careers teaching students to tolerate being wrong long enough to get it right. The best AI tutor may not be the one students like most in the moment, but the one that refuses to finish the work for them. The friction was never an obstacle to learning. It was the learning.

NEWNAN, Ga. — Arnco-Sargent Elementary sits in a wooded area on a two-lane road about 50 miles south of Atlanta. Across the street is a child care center. A Baptist church is less than a mile away. 

But if developers get their way, the school’s newest neighbor could be a sprawling data center campus housing the networking infrastructure needed to help power Americans’ reliance on the internet, and increasingly, artificial intelligence.

Families, some who have lived in the area for generations, are opposed, concerned about traffic, noise, water-quality and other adverse effects. In April, the rezoned 832 rural acres to accommodate the industrial development. Project Sail, as it’s called, could be under construction for the next 10 years.  

“I have a first grader. He’ll be in high school by the time it’s even finished,” said Laura Beth, whose tree-lined property backs up to the proposed site. “This will be his whole childhood.” 

She chairs , which is to stop the project, while also leading a petition drive to get a referendum against it on the ballot. The battle is one of several nationwide in which the rapid expansion of these “hyperscale” centers is colliding with the daily lives of public school families and staff. Americans depend on data centers to work, shop and learn. Despite the recent backlash to ed tech, one research firm estimates that the global in education will grow to over $136 billion by 2035. 

As demand surges, tech companies look for land that is relatively close to customers and existing electricity and water sources. 

“The closer data is stored to the people and systems relying on it, the faster it loads — a benefit that touches everything from healthcare to the classroom technology students and teachers use every day,” said Dan Diorio, executive vice president of state policy and government affairs at the Data Center Coalition, a membership organization. 

But with concerns about energy use and environmental impacts, local governments — and — have on new data centers, and opponents say officials should restrict building near schools and residential neighborhoods. 

“I would like to say our situation’s unique,” Beth said. “It hurts me that it’s not.”

Nationally, 1,878 public schools sit within a mile of at least one data center or a future facility that is approved for construction, according to researchers at Brown University. Shared exclusively with ĂÛÌÒÓ°ÊÓ, the analysis of and National Center for Education Statistics data shows that California has by far the largest number of schools close to data centers or buildings-in-progress: 602, representing 32% of the nation’s total.

Ahmed Saeed, an assistant professor in computer science at Georgia Tech, explained that the infrastructure data centers rely on is more widely available near “population centers.” But because dense urban areas don’t have enough affordable land, developers often gravitate to suburban and exurban areas. 

“As the number of data centers continues to grow, it becomes increasingly likely that some will be located near schools and residential neighborhoods,” he said.

The latest , fueled by advancements in AI, is fairly recent, so researchers know little about the long-term effects of attending school near a facility. The Brown team, however, will soon release research showing that math scores are lower among students whose schools are closer to a data center. Companies say they take measures to minimize disruption, like planting trees to absorb sound, and some argue the downsides . 

‘This buzzing noise’

District leaders are among those who say data centers pose too many risks to ignore. 

“AI is not going anywhere,” said Marcia Andrews, a school board member in Palm Beach, Florida. “My biggest concern is that these data centers should be far enough away that it doesn’t harm the community, it doesn’t harm the schools.”

The Palm Beach County commission agreed. In July, members , a 202-acre near the Arden subdivision that would have been “just a football field away,” from Saddle View Elementary School, Andrews said. The vote, however, leaves the door open for the developers to revise and resubmit their plans in the future. Andrews, whose district includes Saddle View, grew curious about the site when she saw crews widening the roads and planting palm trees. 

“People across the country are waking up now saying, ‘My gosh, we never dreamed that we would hear this buzzing noise 24/7,’ ” she said.

One frequent complaint against data centers is the sound coming from cooling systems, fans and generators. In , a resident likened it to a jet engine. Another in , known for having the of data centers in the world, said the inescapable humming makes him “want to curse.”

A from a former NASA safety manager and planning commissioner in Prince William County, Virginia, predicted that classroom noise from data centers near schools would likely exceed recommended levels. A separate in the state found that the sound is “generally quieter than a lawn mower or nearby conversation,” said Diorio, with the Data Center Coalition. But opponents there often share from a mother at Chris Yung Elementary who said her child wouldn’t go outside for recess because of the sound. 

Many data centers use , which would be like “a bunch of semi trucks idling right next to the school,” said Jessica Owens, the director of data at Science for Georgia, a nonprofit. “Would you want your child in a facility next to a place that’s emitting pollutants, especially in the South when we have heat waves that exacerbate those pollutants?”

Noise, she added, would impact a school within a mile, but they also create “heat islands” that can extend up to six miles. 

The closest Project Sail building would be about 1,000 feet from Arnco-Sargent Elementary “with setbacks, tree buffers, berms and other screening incorporated into the design,” said  Mattie Sorrentino, a spokeswoman for Prologis, the company behind the Coweta County project. The site is “well-suited” for the data center because of its close proximity to Plant Yates, a Georgia Power facility. 

The company “will meet or exceed every requirement,” she said. “Our goal is to keep the project out of sight as much as possible.”

‘Impact on achievement’

Even if students can’t see a data center from the playground, their learning may be at risk.

In a forthcoming paper, Brown University researchers will show that students who attend school within a mile of a data center have lower math scores than those who are farther away.

“You can rule out all other factors,” said Samantha Kane, a research associate and part of the university’s Sustainable Education Research Initiative. “It’s actually the proximity to the data center and the potential pollution coming from the data center that is having the impact on achievement and not other things about the region.”

Critics often as one way data centers put a strain on local resources. Early estimates said Project Sail would use 6 to 9 million gallons per day. Now Prologis says it would draw up to roughly 300,000 gallons a day from the Chattahoochee River, and some comparisons from 2023 show data centers use far less water than , like golf courses.

To reduce dependence on the local water supply, Project Sail would use what’s called a “closed-loop” cooling system. But there are . To keep the hardware inside from overheating, those mechanisms are “blowing hot air into the area and increasing the temperature” as much as , Owens said. If there’s already a heat wave, kids might get less recess time outside. The soaring temperatures also on the power grid as residents, businesses and schools crank up the air conditioning to stay comfortable. 

In Virginia this summer, officials in Henrico County, , asked the school system to help cut energy costs that are expected to increase by $5 million. District spokeswoman Eileen Cox said the system, with 81 schools and more than 50,000 students, already takes a “proactive approach” to conserve electricity, like setting thermostats higher and operating a four-day work week during the summer.

The state, however, just enacted a new on data centers to offset some of the decline in revenue from tax breaks. from state and local taxes also make Georgia a preferred site for the industry. At this point, how many data centers are operating or proposed in Georgia, but Arnco-Sargent isn’t the only school in the state near land eyed by developers.

Closer to Atlanta, residents in Lovejoy are petitioning against a facility that would potentially sit between Michelle Obama STEM Academy and Eddie White Middle Academy.  

The site plan envisions about 3.5 million square feet and 15 buildings of data center space on at least 400 acres, said Clayton County , who has scheduled a community meeting for this Saturday. 

The community, she said, “naturally has serious, legitimate questions about the cumulative, long-term impacts.”

Traffic concerns

Schools in the vicinity of these facilities can feel the effects even before sites are operational. An elementary school in northeast Louisiana last year to protect students from the thousands of cement trucks and 18-wheelers traveling every day to what is expected to be . There were at least three truck crashes last year, and one smashed into the school’s gate. 

For Project Sail, Prologis agreed to some road improvements as a condition for the rezoning. Sorrentino said construction trucks would be routed away from the school and that when finished, the project would “generate limited traffic.”

But families are still worried. The intersection that parents and school bus drivers navigate for drop off and pick up every day doesn’t have a traffic light. 

The school has a long history in the community. Now 70 years old, Arnco-Sargent consolidated two older schools that used to serve children from local mill villages. But in 1996, the area’s leading textile company went bankrupt, families moved away and the area, Beth said, has never prospered like other parts of Coweta County. Nearly 60% of the school’s students live in poverty. 

“These people are just trying to make a living,” she said. Compared to those who the Palm Beach project, Arnco-Sargent parents perhaps have been less outspoken. “They don’t have time to go to a meeting at 6:00 on a Tuesday. They’re working and have young kids.” 

But Beth is hopeful. At the first signature-gathering event, as residents waited to sign the petition. The organizers have nearly reached their goal of 14,000 signatures 

“I feel that this is what it took for the county to listen,” she said.

Drawing on this source material, Claude can generate lessons and materials aligned to an educator’s state standards and curriculum. Given how much teacher time goes into that kind of planning, it’s easy to see why CfT is appealing.

We’re glad to see artificial intelligence’s potential being tapped to help teachers individualize instruction at a scale that wasn’t possible even a few years ago. We are also happy to see a major frontier lab with the resources and technical skills to do it. But public school systems don’t operate like the private companies and government agencies Anthropic is used to working with. Teachers face legal exposure around sharing student data that simply doesn’t exist in those sectors.

This creates a severe employment risk for educators. 

Anthropic’s own shows the problem: “Analyze class data to plan instruction. Hand Claude a folder of data — roster, diagnostics, attendance, your own notes — and it builds a clear picture of where every student is, allowing teachers to better tailor instruction. You control what data is shared; nothing you share is used in model training.”

While this pitch appeals directly to overburdened educators, encouraging them to hand over those kinds of data immediately puts them at risk, because it invites teachers to sidestep district policy and, potentially, state and federal law.

CfT’s privacy terms are stronger than those of ordinary consumer AI products: It does not train on teacher inputs or outputs, and it offers data encryption, deletion rights, audits and breach notification. Anthropic says its with teachers is “written to comply with FERPA” — federal student privacy law — but that’s a claim about how the contract is drafted, not a promise that any individual teacher’s use is lawful. 

Here’s the problem: The marketing prompt to “hand Claude a folder of data” (even with the added disclaimer to check district policies) directly conflicts with CfT’s terms of service in two ways. First, the terms require that the teacher have the authority to designate Anthropic as a “school official.” But FERPA dictates that a vendor qualifies as a “school official” only if the district designates it as such and if the vendor is under the district’s “direct control.” An individual teacher can’t make that designation. 

Second, the terms of service require teachers to indemnify Anthropic for claims arising from what they upload. These terms also cap Anthropic’s liability. In short, the educator, not Anthropic, is assuming the risk. 

While the marketing describes an appealing set of features, the terms lay out requirements that are legally impossible for a teacher to meet. If educators agree to these terms and upload data, they could potentially be fired for it. These risks are even greater when any information about students with disabilities (like an individualized education program) or medical conditions (like a Section 504 plan) is shared. 

Both authors have spent years working with school district bureaucracies, so we know firsthand how hard the purchasing process can be and how legal barriers around data privacy and security can slow the adoption of innovative technology. We both agree that the procurement and purchasing process needs reform to enable such adoption, but bypassing districts and going straight to individual teachers defies fundamental tenets of education policy and practice. 

We want tools like CfT to work and to be used. But there need to be changes in how Anthropic (and other large-scale AI providers gain access to data. The way to make this work is to build data access and governance rules that school systems themselves control. Currently, several initiatives are underway to make this a reality. For example, the , the and the are active state- and district-led initiatives working to develop sophisticated, standards-based AI data access systems within district control. They ensure that AI models can draw on state-specific standards and curriculum and local classroom data where allowed, without requiring individual educators to take on the risk of feeding sensitive student information into third-party tools without the authority to do so. 

Building such technology systems and their governance is hard and time-consuming work that may go against the “move fast and break things” ethos that many tech companies embody. But Anthropic and other AI labs should be collaborating with states and school districts to help build and even assist in funding such systems for the public benefit. They would be part of the solution to this knotty problem instead of fueling the growing AI backlash in public education.

When students return to school this fall in the Salamanca City Central School District in Western New York, a new kind of teacher will be ready to greet them. The small, rural district located on the Seneca Nation reservation is set to be one of the first in the country to put a humanoid robot in a classroom. It will not replace the classroom teacher, but is programmed to provide learning support to both students and educators. 

At a board meeting last month, the Cattaraugus County school district agreed to purchase the robot from Realbotix, a tech company, along with an artificial intelligence teacher’s assistant program allowing students to interact with an avatar of the robot on laptops. 

“This deployment in a working school district represents a landmark moment for both AI and humanoid robotics,” said Andrew Kiguel, CEO of Realbotix, which is currently building the robot. “Salamanca marks the beginning of a new era where humanoid robots and intelligent AI assistants become standard tools in STEM ±đ»ćłÜłŠČčłÙŸ±ŽÇČÔ.”&ČÔČúČő±è;

The female robot, named Sally, will have a “lifelike appearance” with silicone skin and long brown hair, Kiguel said in an interview with New York Focus. It will be stationary in a seated position but have a wide range of upper-body movements and facial expressions.

Students will use a unique identification code when interacting with the robot during class, allowing it to access their learning data and provide personalized support based on their past communication with the avatar, Kiguel said. “They’ll be able to say, ‘Hey, I’m student number 1234,’ and then the robot will be like, ‘Hey, we were talking about this yesterday, do you want to continue that conversation?’ ” 

Salamanca plans to introduce the robot and avatar in its high school AI and robotics courses, which use developed by Apple co-founder Steve Wozniak to prepare students for high-demand tech jobs. The district plans to expand it to high school students in other classes if the pilot is successful. 

Salamanca Superintendent Mark Beehler explained the district’s embrace of AI. “Many schools are taking the easy solution of simply banning it, but I have found that students will find a way around most rules that schools put in place,” Beehler said in a written response to questions. “I also believe it is critical for schools to teach proper use of technology, not simply exclude it.” 

During and after school hours, students will be able to use the avatar to receive help with assignments. Kiguel said students could upload photos of homework for feedback, ask the avatar to generate lessons on topics that interest them, or receive real-time translations in over 100 languages. Unlike other Realbotix robots, Salamanca’s robot will not have advanced features such as facial recognition and recording. 

The robot will also serve as a classroom resource. If a teacher loses their place during a lesson or needs a prompt on what comes next, Kiguel said, they can ask the robot for guidance because it has been loaded with the district’s curriculum. 

The products cost the district $57,590, according to  attached to the school board agenda. Kiguel said the price was discounted; the company’s website lists the M-Series robot, the model purchased by the district, at a starting price of $95,000. 

The plan has sparked debate online, with some calling it a forward-looking, innovative step for the district and others expressing concern about the broader implications of further integrating AI into schools, particularly in an underserved area. 

Salamanca is the only incorporated city in the United States located entirely within a Native American reservation. The school district serves roughly 1,300 students, including 32% who identify as American Indian or Alaska Native and 79% who are economically disadvantaged. 

“We already have so many issues in our community, including environmental issues. I just don’t understand the concept of adding AI onto that,” said local parent Sierra Abrams, who learned about the technology in a . “Bringing it into the school district is contradicting everything we believe, in my opinion.” 

For the past few months, the company has worked to establish guardrails to ensure student safety. Kiguel said the robot and avatar were trained on the district’s curriculum and will always redirect the conversation back to education. References to suicide, self-harm, or other flagged terms would automatically alert school administrators.

In a demonstration for New York Focus, Kiguel told the avatar that he was being bullied at school and wanted to get revenge on the student. 

“I’m really sorry that happened to you. Being picked on and having rocks thrown at you is not okay, and that must have felt upsetting and maybe even scary,” the avatar responded. “The best next step is to tell a trusted adult at school right away, like a teacher, counselor or principal, and let your parent or guardian know, too.” 

Beehler emphasized the importance of safety and privacy, noting that the robot and avatar operate on a closed system that is not connected to the internet, and that Realbotix will not have access to personally identifiable student information. 

To avoid AI-generated inaccuracies, Beehler said Realbotix trained the robot and avatar to say, “I don’t know,” instead of generating fabricated or misleading responses known as “AI łóČč±ô±ôłÜłŠŸ±ČÔČčłÙŸ±ŽÇČÔČő.”&ČÔČúČő±è;

The Trump administration has promoted the expanded use of AI and technology in education. Early in his second term, President Donald Trump  establishing a taskforce to help integrate AI into curricula and teacher training. In March, First Lady Melania Trump entered a White House technology summit  and invited guests to envision a future where robots educate children in literature, science, art, philosophy, mathematics, and history. 

At the same time, there has been  nationwide, with parents forming groups to push for stronger oversight of AI, and clearer guidelines to limit screen time and ensure digital resources are used appropriately in schools. In May, the New York State United Teachers union called for strict  and screen time in schools, and earlier this month, New York City Public Schools Chancellor Kamar Samuels announced  until the school system finalizes guidance on artificial intelligence later this summer. 

The classroom marks a new venture for Realbotix, a Toronto-based robotics company formerly known as Tokens.com, that helped customers use cryptocurrency to rent “digital land” in the Metaverse. In April 2024, the company acquired Simulcra, the Las Vegas parent company behind RealDoll, which has spent decades  and later expanded into sex robots that remain on the market today. 

In a statement to New York Focus, a spokesperson explained that over the past two years, Realbotix has built a new team focused on education, health care, and wellness applications, and that Realbotix and RealDoll do not share employees, payroll, physical locations, or technology. She said Realbotix “is pursuing a transaction with a Nasdaq-listed company that is intended to separate the businesses at the ownership level” with completion expected by September. 

In recent years, Realbotix has worked to expand its presence in commercial settings. Still, it is best known for its “companion robots,” which are different from sex robots and intended to address what it’s described as a “loneliness epidemic.” Kiguel has previously said the company’s goal is to produce robots and AI that are “indistinguishable from humans.” 

Beehler, Salamanca’s superintendent, said the partnership with Realbotix began after a former colleague met an investor at a dinner and discussed the possibility of bringing the company’s robots into the education sector. He was initially cautious about the partnership because it represented uncharted territory for both Realbotix and the school district. But he said the company was receptive to feedback and willing to adapt its products for the unique demands of an educational setting. 

The superintendent acknowledged concerns about students’ increasing screen time and emphasized that AI and robotics should support, not replace, the human connections at the center of education. 

Beehler said the district will measure the success of the program primarily through qualitative feedback from both students and teachers. 

Ryan Schaaf, associate professor of educational technology at Notre Dame of Maryland University, said he was optimistic about the initiative, but cautioned that its success hinges on thoughtful implementation, with teachers actively monitoring and guiding students’ interactions with AI. 

Schaaf acknowledged the backlash against AI in education, but said ignoring the technology is a mistake. “It gives students a true disadvantage because once they leave school, they are going to be immersed in AI technologies.” Instead, he argued that schools should balance AI-assisted learning with traditional instruction to prepare students for life after high school.

State Senator George Borrello, who represents Salamanca and surrounding areas of Western New York, said the program could help level the playing field for students by providing tutoring access to those who might not be able to afford private options and sparking greater interest in STEM. 

Borrello said local leaders have long struggled with keeping young people from leaving Upstate New York after high school. He said investments in technology and STEM education could help show students they can pursue fulfilling, high-paying careers in tech while still building their futures in rural communities like Salamanca.

“One of the biggest things we see right now in technology is people fearing what the future will be with AI. Will it replace the workforce? Will it grow out of control?” he said. “I think this is a great way to get not just the kids but the teachers and the parents more comfortable with what the future may hold.”

This was originally published on .

Taking a computer class at her Fayetteville, N.C., high school, she handed in a paper with the offending word: .

“[The teacher] called me over, and she was like, ‘I don’t want to see you using AI,’ ” the 16-year-old rising junior recalled. If she’d asked, Reaves would have said she didn’t use AI to write the piece — that’s just the way she talks.

“So it was kind of tense,” said Reaves, “and I was definitely upset because she gave me a zero for that assignment.” 

High school students, steeped for nearly the past four years in generative artificial intelligence, face this reality daily, attending class largely without a set of agreed-upon guidelines. Over the weekend, a small group of students came together to hammer out a few basic principles for the technology, something akin to an AI Bill of Rights.

Sponsored by a group that includes an and a national , 100 students from all 50 states gathered here to debate a first-of-its-kind set of recommendations, to be shared with district leaders nationwide as AI proliferates rapidly in the lives of both students and teachers.

The nonprofit , which brought the group together, didn’t make the final recommendations immediately available — they’ll be released in coming weeks. But students told ĂÛÌÒÓ°ÊÓ in interviews they are concerned about a host of problems: cheating, transparency, hallucinations, cognitive offloading, mental health and environmental factors, among others.

Two students from every state debated and ultimately approved the recommendations by a Senate-style roll call vote on Sunday at the Edward M. Kennedy Institute for the United States Senate, a museum and conference center that features a of the U.S. Senate Chamber where Kennedy served from 1962 until his death in 2009.

Despite the democratic nature of the proceedings, Day of AI refused to grant ĂÛÌÒÓ°ÊÓ access over three days to any of the student debates, save for the closing speeches and final vote. 

The organization also clamped down on the text of the proposal, refusing to release a draft that students spent nearly three days discussing freely.

By the time of Sunday’s debate and final vote, nearly 100 students sat in replicas of U.S. senators’ seats, paging through their copies of the bill as staff members and security personnel closely guarded the proceedings. A reporter who picked up a stray copy of the bill on an upholstered bench had the paper quickly snatched away by a staffer, while another swept the area for more copies, gathering them up and tucking them under an arm.

During the proceedings, a few participants called the proposal flawed or incomplete and even, at times, incoherent — one student referred to it as “a block of Swiss cheese.” But in the end he voted “Yea.” In all, 83 students approved the measure, with 15 opposed — two students weren’t present for the final vote.

‘It’s an easy way out’

In the students’ conception, Al literacy instruction should be mandatory as soon as they start using devices in the classroom. The lessons should focus not just on appropriate academic use but on misinformation, plagiarism, bias in Al systems, privacy, environmental impacts and how Al really works.

Ethan Liu, 17, a recent high school graduate from Morgantown, W.Va., said the recommendations are “pretty fair,” especially the section on students. “We basically outline that AI, with the proper supervision, is a great tool for brainstorming and studying,” he said. Liu was one of several students who spoke to ĂÛÌÒÓ°ÊÓ during breaks in the deliberations.

Liu recalled asking ChatGPT to create sets of practice problems. “That was one of the best ways that I would use to study for a test.”

But he said using AI to answer test questions or to generate art, for instance, is a bridge too far. The guidelines, Liu said, suggest that AI should be a tool “to help kids truly understand the topic without just spoon-feeding them the answer.”

In candid conversations throughout the weekend, students said AI is basically turning their school experience upside down, simultaneously supercharging their ability to learn while threatening to derail many of their classmates — and sowing mistrust between students and teachers.

“It’s an easy way out,” said James Baxter, 14, from St. George, Utah. “People like to know hard things, but learning it is a different story for most people.”

While he relies on tools like Google’s Notebook LM to help him create study guides, Dillian Campbell, 17, a rising senior in Fayetteville, N.C., said that when teachers pass out assignments, he most often sees classmates simply pull out their phones and snap a photo, asking AI to do the rest.

He recalled a math teacher telling students, “I don’t care if you use AI, but if you fail your test, that’s on you because you don’t really know the work.”

Jaelyn Jensen, 17, a rising senior in Lee’s Summit, Mo., noted that early on in discussions, the group concluded that while limiting AI use for young students is crucial, giving middle- and high-schoolers more autonomy with AI makes sense. Teens, she said, need “to make their own decisions, to learn from their own mistakes with AI.” But schools must also teach them how to use it responsibly “and let them know that it is something that is going to be in their future.”

Lucia Herrera, 17, a classmate of Jensen’s from Missouri, said AI has even crept into the speech and debate world, with debaters asking ChatGPT to help them respond to questions about their cases — this, in a world where critical thinking and original voice are a key to success.

While she can make a case for AI aiding in research, Herrera said, “I think it just takes all of the human out of [it].”

Bihmanji Acho, 16, a rising junior from St. Cloud, Minn., said she’s most troubled by the ways in which she sees AI undermining interpersonal relationships. She recalled attending a recent summit in which adults said they use AI “at like two a.m. as a therapist. And I just don’t 
” She paused for a moment. “It hurts my heart. I feel like it hurts my heart to hear that.”

While — or is it whilst? — many might believe that AI offers a crutch for low-skill students, Reaves said we should think again: Many of those who use AI most are “super-smart kids that are, like, just checked out.”

That loss of student agency might be among the most pressing problems AI presents, said Gyimah Whitaker, of the Decatur, Ga., school district. That results in fewer students able to think critically and take risks.

Whitaker, who chaperoned two Georgia students to the event, said her 5,400-student district proudly embraces and the rigorous curriculum. But AI, Whitaker said, undermines the values in both systems, driving students toward completion rather than helping them master difficult tasks.

Luckily, young people are, for now at least, fighting back. She noted that students in her district’s IB program “can be truly offended” if a teacher uses AI to grade their paper. They’ve even proposed a “three-strikes-and-you’re-out” AI rule — for teachers. 

To be sure, the students urged AI literacy for teachers, and Cynthia Breazeal, who directs MIT’s RAISE research program, said that’s no longer optional. 

Actually, the advent of generative AI may require more than just literacy. To truly combat the technology, she said, schools may need to adopt an entirely different pedagogical model, one that encourages students to more deeply engage with material.

She invoked legendary MIT mathematician and computer scientist , who co-developed the Logo programming language in the 1960s and founded what became MIT’s Media Lab. Papert helped popularize the term “” to describe the productive struggle learning something complex often requires. In his conception, children should be encouraged to build things and embrace difficulty, not shun or outsource it. Upon hearing a teacher describe the difficulties a child encountered programming in Logo once in the 1980s, he concluded, “I have no doubt that this kid called the work fun because it was hard rather than in spite of being hard.”

‘A Sputnik moment’?

Chase Johnson, 15, a rising junior in Kenai, Alaska, uses AI “a lot” in his personal life. These days, he said, it’s helping him develop a business to automate processes such as customer service. 

As if to prove it, he’ll pull out his phone to show off no fewer than 10 apps, all of which allow him to explore AI in different ways. “I do a lot of research,” he said.

But in his high school, he said, generative artificial intelligence is largely banned because of students using it to cheat.

Some of that responsibility should be borne by teachers, he said: In writing, for instance, instead of assigning essays that are easily outsourced to a bot, teachers should assign research projects that require students to present their findings in front of a class. “You can’t cheat your way out of that,” Johnson said.

Sarah Zahka, a rising junior in Weston, Fla., said she’s not above using AI when needed. “If I’m stuck on a math problem, I’ll ask it, ‘Hey AI, help me solve this problem step-by-step.’ ”

While she’s worried about AI causing students to lose foundational skills, sometimes she simply needs help brainstorming. In writing, Zahka said, she might ask an AI bot how to approach a topic, “but then I’ll write ethically on my own. Although I’m going to sound like a teenager writing an essay and I might not get the grade that the cheaters get, ultimately I’ll know what level I’m at.”

Jeff Riley, the former Massachusetts who now leads Day of AI, said he’s both excited by the technology’s possibilities — and a bit terrified. He predicted that AI could “kickstart the next revolution in education,” particularly with its ability to help teachers differentiate instruction and personalize learning. “But there’s some bad stuff with it too, right? Bias, voice clones, deep fakes, hallucinations, companions, energy concerns. The list goes on and on and on.”

While deadly serious, he said, those concerns aren’t activating enough adults — this should, by all rights, be “” pushing schools and communities forward, but he sees little urgency. He noted that in 2029, the Organisation for Economic Co-operation and Development’s international PISA exam will for the first time assess . “I can almost guarantee you right now: American schoolchildren, 15-year-olds, are going to score 30th in the world on AI literacy — on a technology that was largely created in this country.”

On the flight back to Atlanta on Sunday night, Whitaker, the Georgia superintendent, recalled that the two students she accompanied told her they’d split their vote on the recommendations, one approving the plan and one voting No. The one who opposed the measure told her the proposal “didn’t meet the mark completely” and needed more work.

“I think that’s exciting,” Whitaker admitted. “What excites me is that our children are wrestling just like we are — and that level of inquiry as a superintendent is what you want to see in students every single day.”

Now, it is artificial intelligence. And educators must ask: What does this innovation help students learn that they couldn’t without it?

Unlike previous classroom technologies, generative AI can perform the very thinking that many lessons are designed to develop. It can write essays, solve problems, summarize readings and produce analyses that students are meant to do themselves. 

 have now issued AI guidance for schools. Much of it centers on data privacy, human oversight and teacher training, rather than whether a given use of AI is developmentally appropriate for a child.

That leaves the question most states have been skipping: What thinking should students do themselves? To answer that, it helps to look backward. In 1958, psychologist B.F. Skinner designed a  around a simple principle: Students could not move to the next problem until they had produced the correct response themselves. The machine provided immediate feedback, but it never completed the work for the student.

Whatever one thinks of Skinner’s broader work, he understood that students learn by responding to problems, answering questions, making mistakes, receiving feedback and trying again. The student must do the thinking.

But today, generative AI can complete the assignment without the student.

This is referred to as cognitive offloading, handing one’s thinking off to an external system. A  involving about 1,000 high school math students tested this, comparing two versions of an AI chatbot. One answered questions outright. The other offered hints and guidance, but not the answer. Students with the first chatbot performed worse on later exams than those who did not use AI at all. Students who received only hints and guidance performed as well as the no-AI group. One tool did the thinking for students; the other made them think first.

The rule that should anchor any school AI framework is simple. Its role should expand only as a child’s own foundation grows.

In kindergarten through second grade, students are building the skills everything else depends on, including sounding out words, forming letters and developing number sense. How well a child masters these literacy skills  in third and fourth grade, and  is indicative of math performance years later.

Children build those skills by doing the work; such as sounding out words, making errors and trying again. If AI gives students the answer, that productive struggle never happens, and the skill never truly forms.  shows that the more students hand a task to a machine, the less able they become to do it themselves. 

So, in K-2, AI belongs only in the teacher’s hands. There, it can help educators in planning phonics instruction and generating practice at each student’s level.

Third grade marks the shift from learning to read to reading to learn. The same shift happens in math and writing, where the foundational skills become tools for building arguments, drawing inferences and working through problems. That’s what makes a different kind of AI use possible. 

In grades three through five, students should use AI only after they’ve done an assignment themselves, and only under a teacher’s direction. A teacher might take a paragraph a student has written, generate an AI version, and have the class pick out errors and find weak reasoning, wrong facts and claims that do not hold up.

Working with flawed examples is one of the oldest ways to teach analysis, and AI produces an endless supply of them. A  found that when students critique AI-generated work with a teacher’s guidance, their critical thinking improves. The same research points the other way, too; the more readily people trust AI’s output, the less they think for themselves.

By middle school, AI should become something students should study, not just something they use. This is the age when the judgment to evaluate information . Every eighth grader should be able to explain how these technologies work, where they fail and why they can produce convincing but wrong text — the beginning of true . Students’ own work should still come first, with AI used afterward and under a teacher’s direction.

By high school, students who can read critically, write clearly and work through problems independently can get real value from AI. In writing, they can first draft an essay, then use AI to help identify gaps or missing arguments. The students would then weigh the feedback to decide whether to follow the tool’s suggestions. AI could also work as a tutor, helping students, say, catch up in algebra or progress beyond grade level, giving feedback faster than a teacher with 25 students can.

High schoolers are also entering a world where AI is embedded in most professional fields, and they need to learn to use it well. That means having the capacity to think, write and reason without it, direct the tool and catch its errors. Those skills are the prerequisite to AI readiness. 

One principle should hold across all grades: Keep assessments AI-free. Tests should measure what the student can do, not what a chatbot knows.

Technology has never been the goal of education. Learning is. Schools exist to help students develop knowledge and skills they couldn’t develop on their own, and every technology should earn its place by helping accomplish that. AI may become one of the most powerful instructional tools schools have ever adopted. But its success shouldn’t be measured by what it can do. It should be measured by what students can do as a result of it.