Рубрики NewsSoftwareWTF

Chinese students bypass AI detectors… with artificial intelligence — because they mark real texts as generated

Published by Kateryna Danshyna

It turns out that to prove that your text was not written by artificial intelligence — you just need to ask artificial intelligence for help.

Chinese students have begun to panic as the time for thesis checks approaches, as local universities have stepped up their fight against artificial intelligence and are actively using detectors to identify generated texts. The problem is that sometimes these the tools label texts «created by AI» as actually written by humans — and students are forced to the impossible «to simplify» their work, or even (paradoxically) ask for help from AI to make the text «more human».

Xiaobin, a senior student of the German Philology Department, said Rest of World websiteShe said that a week before her dissertation was due, she received a notification from the university about the mandatory check of all papers with an AI detector — any text that is marked as 30% AI-generated will be rejected.

Xiaobing was not worried, as she had written the 16-page paper herself, using ChatGPT and DeepSeek only to edit a few paragraphs. However, to make sure she was successful, she paid about 70 yuan ($10) on one of the platforms to have the text checked. To her great surprise, half of her dissertation was labeled as having been created by artificial intelligence.

«The whole process seemed absurd to me… I feel like an innocent person being dragged to the gallows,» Xiaobin says.

Currently, there are more than 10 Chinese universities, including the leading Fuzhou, Sichuan and Jiuzhan universities, set a limit of 15-40% of AI-generated text in students’ papers. The requirement is mandatory for graduation theses, and violation may result in a failing grade or even expulsion.

Students complain on social media that they have never or rarely used AI and have gotten similar results to Xiaobin in their tests. Some are trying to simplify their work, turning initially good texts into clumsy and childish onesWhen manual editing fails, young people turn to AI for help. In fact, the university’s decision only encouraged students to use the technology.

It is worth noting that the perfect AI detection tool has not been created yet — Chinese educational institutions mostly rely on detectors developed by the China National Knowledge Infrastructure (CNKI) and technology companies Wanfang Data and Chongqing VIP, which have long been selling plagiarism detection tools. However, some of these platforms also offer plagiarism reduction services using artificial intelligence to help students circumvent the checks, creating a kind of «profit loop».

Currently, there are a few main ways that students use to circumvent the detectors: there are platforms that offer «purely human rewrites» that cost about $100; and actual cheaper chatbots that can replace some words and syntax. Another student, Dede from Fujian Province, paid about 500 yuan ($70) to a tutor who promised manual editing — the score for AI went down, but the text itself became incoherent in places with misinterpreted terms or inappropriate characters.

«It was so obvious that she used an AI tool to edit my paper instead of rewriting it herself,» says Dede, pointing out one mistake: the tutor changed the reference «three knives» (a traditional hair ornament worn by Fujian women) to «three-bladed tools», which, you see, sounds ridiculous.

Another student said that the service used AI to change the word «semiconductor» in her paper to «0.5 conductor».

As for Xiaobing, she took it upon herself to rework her own written text, swapping words or sentences. In the end The trick of changing periods to commas helped: the presence of AI immediately dropped from 50% to 20%

«It feels like you’re being punished for writing too well», — says a girl who eventually successfully passed her dissertation, which was labeled by the university as 2% AI-generated.

Контент сайту призначений для осіб віком від 21 року. Переглядаючи матеріали, ви підтверджуєте свою відповідність віковим обмеженням.

Cуб'єкт у сфері онлайн-медіа; ідентифікатор медіа - R40-06029.