The Korean government’s plan to continue tracking AI-assisted cheating even after the country’s annual college entrance exam has drawn criticism as unfeasible, especially since the effort hinges on cooperation from overseas tech giants. Despite securing pledges of cooperation from four unnamed companies, authorities have yet to determine what data they can obtain or how they will access it.
On Tuesday, the government announced it would analyze internet connection records from telecommunications companies and AI usage logs from global tech firms to identify suspected cheaters as part of its crackdown on cheating during the 2027 College Scholastic Ability Test (CSAT), which took place on Thursday.
“We have secured commitments for active cooperation in providing image capture records, AI service usage logs and information on cloud-based analysis processes,” Deputy Prime Minister and Science and ICT Minister Bae Kyung-hoon said, adding that the cooperation would be “limited to this year’s CSAT.”
President Lee Jae Myung also announced the agreement on X the same day.
“We have agreed with AI companies to obtain records of all instances in which users photographed exam questions and submitted them to AI services during the exam,” he wrote.
However, the details of how the plan will work have yet to be finalized.
The government said it had reached agreements with four companies but declined to identify them. The Ministry of Science and ICT said discussions were still underway over the scope and method of data sharing, including whether companies would provide AI-generated responses and whether the government could obtain records without a warrant.
“Although an agreement to cooperate has been reached, legal issues remain because these are negotiations with private companies,” a ministry official said.

The information technology (IT) industry is less optimistic than the government.
“I’m not sure whether it would be practically possible to identify photos of CSAT questions among the enormous number of queries submitted to AI services and provide only those to the government,” an IT company
Industry observers also point to legal obstacles in the United States, where the Stored Communications Act restricts service providers from voluntarily disclosing the contents of electronic communications to the government. They say companies would likely struggle to share such data voluntarily, even if they had reached an agreement with Korean authorities.
Time is another concern. Obtaining data from overseas companies often takes considerable time, even when authorities submit formal requests for investigative assistance.
“Once the server’s data retention period expires, there is no way to recover the records,” a criminal defense attorney said on condition of anonymity. “International cooperation can take as little as a week, but it often takes one or two months or even longer.”
OpenAI and Anthropic generally delete conversations from their servers within 30 days after users delete them. OpenAI may preserve data for up to 90 days in response to a law enforcement preservation request, but the records could still be deleted if the investigation takes longer.
Domestic AI service providers, meanwhile, have been cooperating with government efforts by restricting certain services during the CSAT.
Services including AI education app Qanda and AI agent app Pickle announced plans to restrict image uploads or temporarily suspend services on Thursday, the day of the exam.
However, overseas AI services such as ChatGPT, Claude and Gemini have not restricted their services during exam hours.
Locally installed AI models — an AI system installed on a user’s computer or mobile device — pose another challenge because they can operate without an internet connection, leaving no network activity records.
Numerous AI models, including Chinese open-
While many experts believe such lightweight models are not yet capable of reliably solving the CSAT’s more difficult questions, their performance is improving rapidly, particularly the Chinese models, posing yet another challenge to the government’s efforts to prevent AI cheating.
