Generative large language models and academic integrity: ethical risks, detection challenges, and Governance in the age of AI

This paper analyzes the impact of generative large language models on academic integrity as a socio-technical issue, rather than a strictly individual problem. The paper does a cross-analysis of 30 high-level Chinese academic articles indexed by SSCI/CSSCI Core, conducts controlled experiments on the generative ability of four major models (ByteDance Doubao, Tencent Yuanbao, Baidu Wenxin Yiyan, and OpenAI Chat-GPT), and does semi-structured interviews with 18 individuals, including editors, professors, integrity officers, publishers, and AI engineers.This paper finds that “surface originality can successfully conceal derivation to circumvent plagiarism and AIGC detection tools such as PaperPass.” The paper finds three themes: surface originality masking derivative content, length-dependent evasion patterns, and institutional responses to evolving evasion. This paper proposes to “enrich the Technology Acceptance Model by perceived ethical risk, develop a concept of a surface-originality evasion ecology.”