Beyond the Prompt: A Dual-Layer Framework for AI-Resistant Assessment in English Literature Education

Abstract English Literature departments face a peculiar version of cheating using Artificial Intelligence, particularly Large Language Models (LLMs). Most of the canon these courses teach from is already public domain, already digitized, and already sitting inside the training data of most major language models. A student does not need to ask an AI to think about Hamlet; the model has, in some sense, already read it a thousand times over. This paper looks at what actions departments may take for it, and where those responses tend to fall apart. Timed online examinations have picked up a set of technical countermeasures over the past year or so: hidden prompt injections buried in the question text, watermarked or visually distorted question images, AI-based proctoring. One documented case from July 2026 caught thirty-two of thirty-five students this way. But these tricks have a short shelf life. Students figure them out quickly, model providers patch around them faster, and proctoring software frequently incorrectly flag attempts which may be genuine. Take-home writing is a harder problem, and probably the more important one for a literature department, since AI detectors themselves are unreliable and biased against certain kinds of writers. Our argument, briefly, is that assignments need to stop rewarding a finished essay and start rewarding the visible work behind it including drafts, in-class discussion tied directly into the prompt, or a short oral defence of the argument. None of this is a permanent fix. It is closer to a set of habits that make AI-generated submissions harder to pass off as one's own, drawn from work in AI security, assessment design, and literary pedagogy that rarely gets read together.