Why edtech AI projects stall
Generative AI fits learning naturally, which is why so many edtech AI features ship fast and then get switched off. They stall on safety for younger learners, trust from teachers and parents, and grading that cannot be defended.
Tutors that just give the answer. A general chatbot completes the homework for the student. Teachers ban it, parents complain, and the feature that was supposed to differentiate the platform becomes a liability.
Off syllabus and invented references. The assistant teaches a method the exam board does not use, or cites a page in a textbook that does not exist. Teachers lose trust after the first example and stop recommending it.
Marks nobody can defend. An AI grade on an essay arrives with no rubric reasoning. When a parent or student challenges it, there is no evidence of how it was reached and no human who checked it.
Children's data in third party models. Learner essays and chat logs are sent to a model provider with default retention settings. Under COPPA, FERPA or the Children's Code, that is a procurement blocker for every school and district.
No safeguarding path. A learner discloses something worrying to the AI tutor. There is no detection, no escalation to a designated safeguarding lead and no log, which is exactly what schools ask about first.
Authoring still takes months. Writing question banks, worked examples and differentiated versions by hand is slow, so new courses launch late and practice content runs thin where learners need it most.