Anti-cheating with AI in exams

PROSECUTION OF CRIMINAL CASE OF USING AI TO CHEAT ON GRADUATION EXAM

While the 2025 high school graduation exam was still taking place, social media spread information that a candidate used AI to help solve problems during the math exam. The Ministry of Public Security at that time immediately investigated and found out the source of the information and determined that 3 candidates at 2 examination councils used AI to cheat during the exam, less than a day after the report was received. All 3 candidates admitted to their wrongdoing.

In early July, Hanoi Police informed that the candidate had secretly brought his phone into the exam room, took pictures of the exam questions, and used two AI applications, Gemini and StudyX, to ask for help in solving them. In addition, at the end of June, Lam Dong Province Police (formerly) also discovered that the candidate used a button camera to directly record the official exam questions for the literature subject, asked his friend outside to use ChatGPT to help him solve the questions, and read the answers.

Anti-cheating with AI in exams and effective anti-cheating solutions - Photo 1.

Not only using technological devices, some candidates have now used AI applications to cheat in important exams - PHOTO: CHATGPT

In the above cases, the police have prosecuted the violators, because the high school graduation exam is listed as a "top secret" state secret.

Sharing at the press conference to conclude the 2025 high school graduation exam, Major General Tran Dinh Chung, Deputy Director of the Department of Internal Political Security (PA03) under the Ministry of Public Security, affirmed that this exam cheating incident occurred on a small scale and there was no leak of questions when the candidates took the exam. This does not affect the security and safety of the entire exam, Mr. Chung emphasized. He also predicted that the use of AI for cheating in the future will become more and more sophisticated.

As AI develops, fraud will spread.

A student in Ho Chi Minh City who is entering grade 12 this year shared that he and his peers are using AI in most subjects, especially when finding information for presentations.

"There are cases where students also use AI to cheat in exams, but not many, because AI only supports social subjects with a lot of theory, but natural subjects often cannot be solved or are solved incorrectly," this student said, adding: "Therefore, I think the more AI develops, the more widespread cheating with AI can be."

The Ministry of Education and Training can coordinate with AI experts - those who understand the capabilities and limitations of AI - to get advice on effective test-making methods in the AI era.

Master Nguyen Gia Hy, lecturer of AI at Swinburne University (Australia)

According to Major General Tran Dinh Chung, this situation requires synchronous solutions to prevent and stop it, such as continuing to apply higher technology to detect cheating using AI or increasing coordination between parties. Mr. Chung also suggested that parties propagate and educate students and related people about the consequences of cheating in exams to prevent this problem from the beginning.

Anti-cheating with AI in exams - Photo 1.

Students use DeepSeek, a popular Chinese AI tool, to solve English questions for the 2025 high school graduation exam - PHOTO: NGOC LONG

PAUSE OR RESTRICT TOOL FEATURES DURING THE EXAM?

Master Bui Manh Hung, founder and operator of the educational application Aiducation, cited the story of many Chinese technology companies suspending services or limiting some features of AI generation tools such as DeepSeek, Doubao... to prevent cheating in the country's university entrance exams.

From there, Mr. Hung suggested that Vietnam could implement similar regulations during the high school graduation exam period. This is an effective method because it helps prevent the problem from the root. Mr. Hung added that the Aiducation education application temporarily turned off AI features related to solving problems during the recent high school graduation exam days to prevent the tool from being abused for cheating purposes.

"I hope that there will soon be a mechanism for technology companies like us to coordinate with the Ministry of Education and Training and relevant agencies, so that we can simultaneously suspend the provision of related AI services to ensure the honesty and seriousness of the exam," Mr. Hung confided.

However, Master Nguyen Gia Hy, an AI lecturer at Swinburne University (Australia), co-founder and CEO of two AI companies in Vietnam, SkillPixel and AIFicient, said that banning "highly effective" AI tools is only suitable for countries with strong domestic AI products such as the US and China. Because in Vietnam, if a student wants to cheat, he or she will mainly use foreign AI tools instead of domestic products.

"China effectively implemented this solution because they blocked almost all US technology products, including ChatGPT. If we block the IPs of those tools during the exam, we can limit the possibility of cheating better, but it will greatly affect other users because they cannot access it either. And even if we block the IPs, there is still a possibility of bypassing if the candidate uses VPN software to change the region," Mr. Hy shared.

If foreign AI companies are required to block the exercise solving feature, they likely will not do so because it is not under the control of the Vietnamese government, Mr. Hy added.

Anti-cheating with AI in exams - Photo 2.

A post on the AI platform StudyX matched a question in the high school graduation math exam on June 26 - PHOTO: SCREENSHOT

CHANGE NEEDS TO START WITH THE EXAM

From the above reality, Master Hy proposed a 3-level action model. At the lowest level is the awareness of the PhD, we need to promote more propaganda and ethics education when using AI. In addition, it is necessary to determine specific penalties and strictly handle cases of fraud using AI to set an example. "Although we cannot change the intentions of stubborn PhDs who cheat, but for PhDs who are on the verge of violation, proper propaganda and education will bring great results," Master Hy stated his opinion.

The next level is to redesign the way questions are asked, because multiple choice questions are "very good conditions" for AI to solve problems with high accuracy. On the contrary, if questions are asked with essays, analysis and situation handling, "it will greatly limit AI's capabilities," Mr. Hy stated his opinion. "The Ministry of Education and Training can coordinate with AI experts - those who understand the capabilities and limitations of AI - to get advice on effective question-making methods in the AI era," Mr. Hy suggested.

"Like the university I am teaching in Australia, students are allowed to take the final exam at home, but they will not get high marks if they use ChatGPT to solve the questions because the questions are designed to handle situations. They need to understand the lesson clearly and know how to apply the content and knowledge they have learned to the test. At this time, even with support tools, teachers still need to really understand the lesson to be able to use the tool to ask questions," Mr. Hy added.

The final level is to enhance monitoring capacity with technology, because if AI is a tool to support cheating, it can also become an anti-cheating tool. It is necessary to pilot the application of intelligent monitoring systems at exam sites such as AI cameras that can detect abnormal behavior of candidates during the test. This is a very popular application in the field of computer vision of AI, Master Hy informed.

Regarding this issue, the 12th grade student in Ho Chi Minh City mentioned above said that if AI does not flourish, the fairness and seriousness of the high school graduation exam will be affected. Because during the review process, many teachers gave the class "anti-AI" exercises - questions that teachers themselves made up or took from "heaven and earth" sources that are not available on the internet and make AI "mistakenly solve". These questions, if applied in the exam, will make the use of AI useless, according to the student.

The world has a headache because of AI fraud

Not only Vietnam, developed countries are also facing many difficulties in dealing with AI.

A survey by The Guardian published in June showed that nearly 7,000 university students in the UK were found to be cheating using AI, such as ChatGPT, in the 2023-2024 academic year. This means that for every 1,000 students in the UK, there were 5.1 cases of AI cheating, a significant increase compared to the figure of 1.6/1,000 in the 2022-2023 academic year.

In Singapore, the number of students caught plagiarizing or submitting papers generated by AI tools is still low, but universities say it is not a common phenomenon.

Meanwhile, an investigation by Nikkei Asia found that many research papers from 14 universities in countries such as Japan, South Korea, China, and the US had hidden prompts inserted so that if AI was used, reviewers would receive only positive comments. These prompts were often highlighted in white or the text was set to be extremely small so that it could not be seen with the naked eye.

Source: https://thanhnien.vn/chong-gian-lan-bang-ai-trong-thi-cu-185250710195249068.htm