Beyond the Binary: Building Process‑based Academic Integrity — A Single-concept Case Study of AI-detection Evasion
The rapid inclusion of Generative Artificial Intelligence (GenAI) in higher education has triggered a severe regulatory crisis, because traditional text-matching plagiarism policies fail to deal with machine-generated content. Although the University Grants Commission (UGC) of India very recently updated its guidelines and categorised unacknowledged AI use as plagiarism, the formal gazette amendments are still in the administrative pipeline. This policy is based primarily on software-driven AI detection and assumes AI-generated text to be static and easily identifiable. This paper presents a preliminary, single-concept case study across three independent generation trials showing how commercial AI detection tools like ZeroGPT, QuillBot, iThenticate, Copyleaks can be vulnerable to structured prompt engineering. Using a ‘Progressive Adversarial Prompting Pipeline’ across ChatGPT, Gemini, and Claude at eight experimental levels reveals significant detection degradation. Specifically, techniques such as syntactic asymmetry, lexical subversion, and stylometric emulation cause marked tool divergence, enabling prompt-engineered text to bypass detection. As a result, depending entirely on AI-detector percentages creates a contradictory loophole where prompt-literate students evade AI detection while users of basic prompts are penalised. To address this, the study introduces a process-based digital integrity framework for the higher education system by re-framing the evaluation paradigm from binary software detection to cognitive verification through prompt versioning logs, continuous milestone tracking, rigorous oral viva-voce examinations, and bibliographic DOI authentication to combat LLM-generated ghost references.