Every AI lab publishes research documenting the real limits of its own models, hallucination rates, contaminated benchmarks, and the gap between what a model claims to reason through and what it actually computes. Almost none of that research reaches the school and district leaders now approving AI tools for classrooms and administration. This session walks through ten specific, sourced findings from Anthropic, OpenAI, DeepMind, and peer reviewed research, then translates each into a concrete question education leaders should ask before trusting an AI tool with student data, grading, or instructional decisions. This is a literacy briefing, not a vendor pitch.