Most claims about AI in education lack the evidence to back them.

Universities are quickly adopting agentic artificial intelligence, which means AI that can plan, use tools, remember details, or act on its own. However, rules for using it safely are lagging behind real proof that it works. Researchers studied this problem by examining fifteen reviews, one preprint, and selected studies. They checked how people actually use and supervise these tools. They found that the science is messy. The clearest proof only shows small technical fixes, like giving more accurate feedback or making fewer factual errors. Almost no solid evidence proves that these tools help students truly learn over time, make education fairer, cut down workloads, or improve entire schools. Most tests were brief and run in just one place without formal supervision. People also mistake the feeling of an AI having agency for actual system autonomy. Basically, big claims about safe, independent AI in colleges are not yet backed up by real evidence.

Why it matters

Universities are starting to use AI systems that can plan, remember, and act on their own. But this review shows that the research behind these systems is thin—most studies are short, small, and don't check if students actually learn better or if the systems are safe. Without stronger evidence, schools risk deploying tools that don't work or cause harm, especially to students who need the most support.

Questions answered