A private school built around an AI tutor is expanding to 50 campuses. Its students are 10.

Two hours after the morning bell at Alpha School, the AI tutor has already covered the day's math. Students move on to an afternoon of coding, entrepreneurship, and public speaking. The school calls the second half "life skills." The rest of the country is not yet sure what to call the first half.

Alpha, a private K-12 company, has built its school day around an AI tutor that delivers the academic core in roughly two hours each morning. The company is now expanding fast. This fall it is opening 27 new campuses, growing from roughly 23 to about 50 U.S. locations, with tuition that runs $40,000 to $75,000 a year depending on the campus (Scientific American).

The question the company is not yet answering is whether the model works outside the early campuses that built it.

The pitch is unusually explicit about what the AI is for. Carl Hendrick, Alpha's senior learning scientist, told Scientific American the system is "trained on many examples and many misconceptions" and compared it to a self-driving car, with one difference. The "cars" have 10-year-olds in the backseat. That comparison tells a smart reader how the engineers think about the problem. It does not tell the reader whether the students are learning more.

The evidence Alpha has shared publicly is the kind that supports a marketing claim, not a causal one. Alpha has released NWEA MAP math and reading scores but not the underlying data. McEachin said the released scores may reflect who is attending, not the experience. The MAP reports, McEachin added, were not designed to support causal inferences about a specific school. That distinction is the gap the model has to close before a public system can take it seriously.

The wider research base on AI tutoring is encouraging in one place and limited in another. A 2025 randomized experiment at Harvard in an introductory physics course found median learning gains more than twice as high for students using an adaptive AI tutoring model compared with in-class active learning (Scientific American). A separate 2025 literature review of 28 studies covering roughly 5,000 K-12 students found generally positive effects from AI intelligent tutoring systems, but the advantage shrank to nearly nothing when the comparison was non-intelligent tutoring systems that also used active learning. The Harvard result is encouraging. The literature review is a warning against treating the format as the intervention.

Goldhaber said the rollout speed and the evidence base are out of step. Gerald LeTendre of Penn State raised a different question: if the model is being trained on a self-selecting cohort of families who can pay $40,000 to $75,000 a year, what does it learn to teach?

That question is no longer hypothetical. Alpha is testing versions of its model in public schools: two in Houston and one in Springfield, Massachusetts, with a much broader student mix. The company has not said how it is adjusting the model for those classrooms or what data it is sharing with the districts. A 404 Media investigation this year reported faulty lessons in Alpha's system; a former employee told the outlet, "Students were being treated like guinea pigs." Alpha spokesperson Anna Davlantes has disputed that characterization.

The public-school context is what turns the expansion from a private experiment into a national question. About one in five U.S. students is chronically absent. Reading and math scores have been declining across the country for a decade (Scientific American). Alpha's founder, Joe Liemandt, has said he wants to "reach a billion kids." The country is not waiting for him. The question for public districts is whether a model validated in private, premium-priced classrooms is the one they should be evaluating while their own baseline worsens.

The 27 new campuses open this fall. The independent validation that would let a public school system take the model seriously has not arrived.