A nonprofit launched an AI tutor, then charged for wider access

The classroom note

In March 2023, Sal Khan told public school administrators about an experimental tool built on GPT-4. Reaction, he wrote in the announcement, was strongly positive; one administrator said the demonstration "aligns with our vision of creating thinkers." Khan Academy opened Khan Labs, a testing space limited to district partners, and invited "a limited number of teachers, students, and donors" to try Khanmigo as a tutor and teaching assistant. For a teacher already using Khan Academy's free courses, this was a curated pilot layered on top of a platform districts already used, not a public launch.

What the evidence says

The announcement is Khan Academy's own account of its own product, not an independent evaluation, and reports no measured learning outcome. It does state, in the founder's words, a limitation: "AI makes mistakes," including in maths, and "AI can still 'hallucinate.'" The same post carries a dated update, added 17 November 2023: Khanmigo becomes available outside the pilot for "$4/month or $44/year" to users "18 or older" living in the United States. The current product page, retrieved 16 September 2026, repeats the same age and country rule and describes Khanmigo as something that "doesn't just give answers" but "guides learners to find the answer themselves." A small pilot and an open subscription are not the same evidence base, even where the name has not changed.

The implementation question

The pricing update answers a family's cost question: an adult account costs a few dollars a month, with children under 18 added as dependents. It does not answer a school's cost or training question, because the pilot described was an invited district arrangement, not the consumer subscription. The Socratic constraint, that Khanmigo "does not, however, give students answers," is enforced through prompting, and Khan Academy does not claim it is unbreakable. This is an editorial point, not one the sources make: a guided-tutor chatbot instructed not to answer directly can often still be persuaded to, if a student rephrases the request as a check on their reasoning rather than a request for the answer.

What holds and what fails

The description holds where the sources are strongest: a small, invited pilot, run alongside classwork, with Khan Academy itself flagging maths errors and hallucination as known risks. It is weaker wherever a reader extends "guides rather than answers" into a guarantee of academic integrity, or assumes the pilot's supervision travels automatically into a subscription used unsupervised at home, which the cited pages do not describe.

• Is a school looking at the invited pilot Khan Academy describes, or the $4-a-month consumer product, and are the guardrails the same in both?
• What happens when a student rephrases a request to route around the "no direct answers" design?
• Does Khan Academy's own admission of maths errors and hallucination change how a teacher should read a demonstration?

Read together, the two documents describe a pilot that became a paid product without a published account of what changed in between, which is the gap a school should ask about before treating Khanmigo as a single, stable thing.