MagicSchool AI

The privacy posture is the strongest in this K-12 group and is worth teaching as a benchmark: it states plainly that it does not use personal information, including children's, to train models, binds its subproviders to the same, and operates zero data retention with them. It addresses FERPA explicitly and the free tier is substantial.

Having students evaluate a generated preschool lesson plan against developmentally appropriate practice and document where the output violates it. Giving students a concrete vendor privacy policy to audit — age definitions, who consents, training commitments — as a professional skill before they run a classroom themselves.

The consent structure matters for future educators: it defines 'child' as under 13 and relies on the school, not the parent, to consent, so parents cannot delete or amend their child's data directly. Generated lesson plans are formulaic and frequently misjudge developmental appropriateness, recommending activities above or below the stated age band — and novice teachers are least equipped to catch that.