The North Carolina General Assembly is moving a bill that would spend more than $10 million in recurring state money every year to put an AI chatbot in front of students in grades six through twelve. No competitive bidding. No independent review of the evidence. No teachers consulted during drafting. Just a contract written into statute for a single named vendor.

The company is Khan Academy. The tool is called Khanmigo. Senate Bill 1006 — written by Sens. Michael Lee, Kevin Corbin, and Brad Overcash, the same Republicans who chair the committees it must pass through — could put it in your child’s classroom before the school year starts in August.

Before that happens, it’s worth asking what we actually know about this tool. The answer is less reassuring than supporters of the bill have suggested.

At its core, Khanmigo is a chatbot. It runs on GPT-4, the same OpenAI technology that originally powered ChatGPT, though OpenAI has since moved on to newer models. It sits inside Khan Academy’s existing website and tutors students by asking guiding questions rather than providing direct answers. The bill calls it the AI Instructional Support Program, designed to help students with “lesson comprehension.” It also generates lesson plans and assessments for teachers.

What it actually does in classrooms is another matter.

What the Evidence Actually Shows

The actual research on Khanmigo is thin, and what exists is not encouraging. The most directly relevant peer-reviewed study, published in 2025, found no statistically significant differences in learning outcomes between students using Khanmigo and students using Google or traditional paper-based instruction. The researchers concluded that while AI-powered tutoring systems can support learning, their immediate impact on outcomes may be no different from conventional methods. That’s the best independent evidence available.

Khan Academy’s own leadership has been more candid than the legislators pushing this bill. In an April interview with Chalkbeat, Khan himself admitted the tool had failed to reach most of the students who had access to it. “For a lot of students, it was a non-event,” he said. “They just didn’t use it much.” The numbers bear that out: only 15 percent of students with access to Khanmigo use it regularly, despite more than 100 million total interactions since it launched in 2023. Khan Academy’s chief learning officer confirmed that year that efficacy studies were still underway — meaning the data on whether the tool actually improves student outcomes didn’t exist yet when North Carolina was drafting this legislation.

Now, the company is rebuilding Khanmigo from scratch for a summer 2026 release. North Carolina lawmakers are moving to spend $10 million a year on a product that the company’s own founder acknowledged most students rarely used and that Khan Academy is now replacing. Paying recurring state dollars to be someone’s beta tester is not a sound education policy.

There’s also the question of who evaluates whether any of this works. The bill requires an evaluation of Khanmigo’s effectiveness, and it assigns that job to the Office of Learning Research at the NC Collaboratory — a body created by and answerable to the General Assembly. In other words, the legislature that is mandating the contract also controls the organization charged with determining whether the contract was a good idea. The report isn’t due until April 2028, nearly two full school years after districts begin spending public money on a tool whose long-term effectiveness has yet to be established. By the time that report lands, the contract will already have been renewed at least once and the money will already be gone.

The Socratic Flaw in Real Classrooms

Khanmigo’s defining feature is its refusal to give direct answers. The idea is Socratic: guide students to understanding through questions rather than just handing them the solution. In theory that’s sound pedagogy. In practice, with real students in real classrooms, it falls apart in a specific and predictable way.

Teachers who observed students using Khanmigo in a Michigan classroom pilot reported something that should give North Carolina’s legislators pause. Students struggled to use the tool not because it was too hard, but because they didn’t know what question to ask. They didn’t know what they didn’t know.

A skilled teacher navigates that problem constantly. They read the room. They recognize when a student is lost versus when a student is thinking. They shift their approach mid-lesson based on what they’re seeing. A chatbot applies the same method uniformly, to every student, regardless of whether it’s working. The students who most need someone to meet them where they are get a tool that keeps asking them questions they can’t answer.

Khanmigo can be rebuilt from the ground up, but it will still be a chatbot confronting the same problem: a child who doesn’t know what to ask.

There is a growing body of research on what happens to students who rely heavily on AI tools for their learning, and it is not a picture that should make a parent comfortable.

A 2024 systematic review found that over-reliance on AI dialogue systems was linked to reduced critical thinking, diminished analytical reasoning, and decreased creativity in students, with researchers warning that when students lean too heavily on AI for answers, they risk losing the capacity for independent thought.

These academic warnings align with what students themselves are seeing: as of December 2025, according to a RAND Corporation survey of more than 1,200 students, 67 percent said they believed that the more students use AI for schoolwork, the more it harms their critical thinking skills, a number that had risen more than 10 percentage points in less than a year. The students being asked to use this tool are more skeptical of it than the lawmakers mandating it.

Learning is a social act. Decades of research on child development are consistent on this point: meaningful interactions with teachers and peers don’t just make school more pleasant. They are how children develop. They are how students learn to navigate disagreement, to persist through confusion, to ask for help, to trust an adult who knows more than they do. None of which stopped the same Republican majority now pushing a $10 million AI instructional support program from cutting nearly 7,000 teaching assistant positions over the past decade and never restoring them. Now, the legislature is proposing to spend $10 million annually on a software-based instructional support system instead.

Nobody knows yet what it means for a child’s development to spend significant instructional time interacting with a machine rather than a person. The research on the long-term effects of AI chatbot use on K-12 students’ relationships, mental health, and social growth simply does not exist. We would be running that experiment on North Carolina’s children, at scale, starting in August. And yet the people best positioned to raise these concerns were never asked. No North Carolina teachers were asked to testify before the education committee about Khanmigo.

Sen. Michael Lee, the bill’s primary sponsor, defended the no-bid contract by saying state officials had spent years evaluating the platform, pointing to meetings between Khan Academy founder Sal Khan and North Carolina officials. He did not specify who was in those meetings, when they took place, or what was discussed. Sen. Sophia Chitlik (D-Durham) wasn’t satisfied. “I’m struggling to find the evidence that shows that this is a better investment than, say, $10 million in instructional assistance,” she said.

When the research base is thin and the evidence is contested, the people closest to students are the ones most likely to ask hard questions. That may be exactly why they weren’t consulted.

North Carolina has seen this movie before. We saw it in 2019 when the state bypassed its own evaluation process to push a controversial, no-bid contract for the Istation reading program—leaving districts locked into a product educators didn’t ask for through the same closed-door maneuverings.

None of this required an elaborate alternative policy agenda. A competitive bidding process, an independent review of the evidence before the contract rather than after, a genuine pilot in a handful of willing districts before any statewide mandate, and educators at the table from the beginning. These are not radical demands. Neither is the question Sen. Chitlik raised in committee and nobody answered: what would $10 million in classroom aides, reading specialists, or one-on-one tutoring do for the students this bill claims to be helping?

They are the minimum that students and taxpayers should expect before $10 million in recurring public money goes to a single vendor, named in statute, for a tool whose effectiveness remains unproven and whose long-term impact is still being studied.

The question before lawmakers is not whether artificial intelligence belongs in education. It is whether North Carolina should commit $10 million a year to a single AI product before independent evidence, educator input, and transparent procurement have occurred. Those are questions that should be answered before August, not after.

Senate Bill 1006 hasn’t passed yet. If you’re a parent, teacher, or taxpayer with concerns about this proposal, now is the time to contact your legislators. Ask them to slow down, bring educators to the table, and answer those questions before committing $10 million in recurring public funds to a statewide AI rollout.