Students who trust AI assistants stop checking their work — unless they understand how AI actually fails

A study of 432 students solving programming problems found that higher trust in an AI assistant correlated with lower-quality decision-making: students accepted wrong answers and rejected correct ones more often. Students with stronger AI literacy and higher intellectual curiosity reversed this pattern, treating the AI's suggestions as information to evaluate rather than authority to follow.

As educators start embedding AI tools into classrooms, the risk isn't that students won't use them — it's that they'll trust them uncritically, especially the students most likely to need scaffolding. The finding cuts against the assumption that access to an AI tutor automatically helps learning; it turns out, students need to be taught to think about when an AI is probably wrong before the tool becomes useful rather than a shortcut that hides gaps in understanding. What changes in practice is straightforward: schools rolling out AI tutors will need to teach students how these systems fail, not just how to use them.