Schools Have an AI Strategy. Do They Have an Assessment Strategy?

AI policies are arriving quickly in education, but assessment may not be changing at the same pace. In a 2026 Pew Research Center survey, 54% of American teenagers said they had used chatbots for schoolwork, while 59% said AI cheating happens at least somewhat often at their schools. This September, New York City placed a one-year moratorium on student-facing generative AI from 2-K through eighth grade, affecting nearly 600,000 students. MIT, meanwhile, is asking instructors to define more clearly when AI should be prohibited, permitted, or required. Schools and universities are drawing boundaries around AI. But there is something deeper here worth paying attention to. If the tools students use to learn have changed this dramatically, should the way we assess learning remain largely the same?

The tension becomes particularly visible when work completed outside the classroom is used as evidence of mastery. At Brown University, economics professor Roberto Serrano confronted this after a take-home midterm in a course he had taught for nearly two decades produced an average of 96%, compared with historical averages ranging from 65% to 80%. Suspecting widespread AI use, he returned the final examination to the classroom, where reported performance fell sharply. One course cannot establish a broader trend, but it exposes a problem schools at every level now have to confront.

Homework and take-home assignments have traditionally been asked to do two jobs: help students learn and demonstrate that they have learned.

AI is making the distinction between those two jobs impossible to ignore.

Perhaps the response is not to keep AI away from homework, but to become much clearer about what homework is meant to accomplish. A student who leaves class with gaps can now ask AI for another explanation, recover a missing prerequisite, explore details that were missed, generate additional examples, practice at different levels of difficulty, and test their own understanding. Used thoughtfully, AI could make homework a richer learning environment and reduce how much a student's opportunity to understand depends on whether everything was captured perfectly during one class period. Schools could give students clear learning objectives and greater responsibility for reaching them: here is what you are expected to know, and here are the resources available to help you learn it. The teacher, textbook, class materials, reliable outside sources, and, where appropriate, AI can all contribute to that process. If homework becomes more explicitly about building mastery, perhaps we should also be willing to let it carry less weight as evidence that mastery has occurred.

That evidence can move closer to the classroom. Supervised classwork, short quizzes, tests, oral checks, and other independent assessments can carry greater weight, while the assessments themselves become better at revealing whether knowledge transfers beyond familiar examples. A familiar concept can appear in an unfamiliar context. Students can decide which information matters, diagnose an incorrect solution, connect ideas previously taught separately, or explain why one answer is better supported than another. There is another side of AI that deserves much more attention too: teachers have access to it. AI can help educators generate larger question banks, create multiple versions of an assessment, surface common misconceptions, and develop new applications of material students have already studied. Teachers still need to review those questions for accuracy, fairness, alignment, and difficulty, but the technology that has complicated traditional assessment could also give educators the means to make assessment more thoughtful, varied, and revealing.

Perhaps this is the opportunity hidden inside the disruption. Schools need AI policies around age, privacy, academic integrity, and appropriate use, but

an AI policy is not the same thing as an assessment strategy.

We may need to distinguish much more deliberately between learning with assistance and demonstrating independent mastery. When the purpose is learning, AI can help students explore, practice, question, and close gaps. When the purpose is verification, students need opportunities to show what remained after the assistance is removed. And teachers can use the same technology not simply to manage AI use, but to push the boundaries of what good assessment can reveal. The goal does not have to be more AI or less AI. It can be greater discernment about where it belongs. Perhaps the question, then, is no longer simply how much AI we should allow in education, but

where should AI expand learning, and where should students still be expected to stand on their own?