On June 3, 2026, Chinese media Yicai reported on the reality of students paying for their own checks to lower the AI rate and intentionally rewriting their graduation theses in a crude manner, which arose from the introduction of AI checks for graduation theses.

The article introduced that many universities in China have introduced generative AI check policies for graduation theses and set the alert line for the AI rate at 20-40%. It reported that students are busy with revision work because exceeding this figure affects anonymous thesis reviews and oral examinations.

It then introduced an episode about Zhou (JOU), a fourth-year university student in Chongqing City, who, because the AI rate of a thesis he wrote himself was judged to be 80% in a pre-submission check, revised it until late at night using a free AI tool the day before the deadline, and barely lowered it to under 20% to obtain defense qualification. However, when he revised it later under the instruction of his supervisor, it exceeded 20% again.

The article mentioned doubts about the reliability of checkers as the background for the problem of "the AI rate that should have decreased rising again." When a reporter checked a 100% AI-generated text using DeepSeek, it showed a large discrepancy in the results, with CNKI judging it 0% and VIP judging it 55.71%.

It also pointed out as a major problem that costs increase because repeated checks are unavoidable despite the questionable reliability. It reported on the unhealthy structure where companies providing checkers profit and even offer rewriting services to lower the AI rate, and highlighted the case of graduate student A Qiu (ARCHIU), who had to repeat checks 7 times and spend a total of 780 yuan (approximately 18,000 yen) to clear the AI rate standard.

Furthermore, it introduced that a contradiction has arisen where students, in order to lower the AI rate, replace technical terms with cheap expressions or intentionally disrupt the arrangement of well-structured sentences, thus deliberately rewriting their theses into lower quality ones.

The article reported that Professor Zhao Bin (ZHAO BIN) of Fudan University pointed out the phenomenon that "the more sophisticated a sentence is, the higher its AI rate, and the more logically confused a sentence is, the more likely it is to be judged as human," and then suggested that universities should not blindly set red lines but clarify the boundaries and norms for use.

It also reported that a relevant academic society, commissioned by the Chinese Ministry of Education, announced guidelines along this direction and proposed a shift from reliance on technical checks to an evaluation that emphasizes students' thought processes. (Edited and translated by Kawashiri)