With increasing workloads, many educators are turning to AI to generate classroom materials quickly and efficiently. AI-created reading texts, worksheets, and listening activities offer level-appropriate content packed with target vocabulary and grammar, spanning topics from SDGs and nursing to the latest K-pop trends. But how accurate and reliable are these materials? Could they even pose risks to students' learning? Cognitive phenomena such as cryptomnesia, source confusion, and the illusory truth effect demonstrate how exposure to misinformation can lead individuals to unknowingly adopt false knowledge as fact. If AI-generated content contains errors, could ESL teachers inadvertently reinforce misconceptions in their students? This study examines the accuracy and educational value of AI-generated materials by consulting subject matter experts—including cinephiles, medical doctors, and a high court judge—who evaluated reading activities produced by three different AI models to rate the accuracy, content and educational value of the texts. Errors were identified, categorized (unimportant, bizarre, or potentially harmful), and analyzed by topic and AI model. This presentation provides essential insights into the risks and limitations of AI-generated teaching materials, helping educators make informed decisions about their use in the classroom.