A new assessment approach argues institutions should grade what students understand rather than what they can submit with AI support. The proposed framework, described as “Grade the Human, Not the Homework,” shifts grading from submission quality to students’ ability to explain, modify, troubleshoot, or justify what they turned in. In the described implementation, the instructor used in-class validation tasks and short quizzes after due dates. Students were no longer given automatic credit for completed programming assignments; instead, they had to demonstrate competency through modifications that matched the original work. The article reports a measurable pattern: roughly a third of students could not complete simple code modifications during the allotted class time despite having submitted correct code, suggesting that submission alone masked learning gaps. The framework aims to reduce pressure on faculty to act as AI detectors while aligning grades with demonstrable skill.