With the advent of generative AI, teachers are faced with the difficult choice of trusting advanced technologies such as Large Language Models (e.g., Chat-GPT) to take advantage of their transformative potential. Trusting educational AI is getting harder as these systems become less transparent, less predictable, and raise equity concerns. In the workshop, teachers attempt to break the AI grader by giving it inputs that are likely to confuse it. Despite being simple changes to input (e.g., swapping a word with its synonyms), teachers will see how they easily flip the AI grade. Teachers will also have the option to come up with new creative ways to break the AI grader.