The Mismatch Vulnerability: Preserving Academic Integrity with LLMs in Education

Large language models (LLMs) such as ChatGPT are transforming higher education, creating opportunities for learning while challenging assessment and academic integrity. In this paper I develop the “mismatch vulnerability,” a design flaw in assignments that occurs when students can pass the assignment without exercising the skills or capacities it is intended to assess, and I argue that instructors should test and redesign assignments with this vulnerability in mind. In these cases, grades become false positives, signaling that learning objectives have been met when they have not. I discuss how this vulnerability extends beyond intentional dishonesty and is intensified by AI tools that can produce work meeting grading criteria through the use of alternative skills, such as prompt engineering. I also introduce the “calculator effect” to explain why students may choose to rely on AI in ways that bypass educationally valuable processes. Finally, I propose a step-wise approach for identifying and addressing mismatch vulnerabilities, including testing assignments with LLMs, to preserve meaningful grades while supporting responsible AI use in courses.