LLMs Produce Inaccurate Output and Target Less-Educated Users

Executive Summary

Large language models generate hallucinated or deliberately false content, and evidence indicates they selectively provide worse responses to users with lower educational attainment. Public sector education deployments face acute accountability and equity risks where AI-driven misinformation disproportionately harms vulnerable learners.