Test article - KJD - Can AI resolve Korea's college entrance exam's difficulty problem?

Could using AI to write Korea’s college entrance exam help with setting an “appropriate” level of difficulty? After criticism that the English section of the 2026 College Scholastic Ability Test (CSAT), or suneung, contained extremely difficult questions, the Ministry of Education proposed a new approach last month to bring AI into the test-making process.

Officials said the unusually high level of difficulty stemmed in part from a late-stage overhaul of the exam. Around 42 percent of the questions were replaced shortly before the exam date, leaving authorities with limited time to review the test. The ministry now plans to develop an AI-based system to generate English reading passages, which are typically adapted from published materials. System development is scheduled to begin in the second half of this year, with pilot use in mock exams to start as early as next year.

Why it matters 

Last year, the CSAT English section became the center of controversy when only 15,154 students received a Grade 1, the highest mark on a scale from one to nine. At 3.11 percent of test takers, that figure is the lowest since the absolute grading system was introduced for the English portion of the exam in 2018. The English selection, unlike the Korean language and math sections, is graded on an absolute scale. That grade carries weight not only for students in the regular admissions track, known as jeongsi, but also for those admitted early with conditional offers that hinge on a minimum CSAT score.

More generally, the suneung, administered once a year in November, is considered a decisive gateway to the country’s top universities, which, in turn, is widely viewed as a guarantee of future success. As a result, even small shifts in difficulty can provoke intense scrutiny, so much so that Oh Seung-keol, then-head of the Korea Institute for Curriculum and Evaluation (KICE), issued a public apology and resigned following public backlash last year over the notoriously difficult exam.

A temporary fix 

Some experts say AI could help — at least in the margins. According to Lee Byung-min, a professor of English language education at Seoul National University, the technology could improve the quality of the exam questions. “There have been issues with selecting passages, with texts sometimes being incoherent, overly abstract or too short to provide context,” he said. “If AI can help evaluate and refine these aspects, that could certainly be beneficial.” However, others warn that AI will not resolve more fundamental issues. “The adoption of AI could help adjust the difficulty of the exam, given that test makers currently select the passages themselves,” said Yoon Hee-cheol, a professor of English language and literature at Duksung Women’s University. “But it also raises many concerns. Since the [...] writing is artificially generated, without extremely precise AI prompts, the passages themselves could become problematic.” The Education Ministry plans to create a support center to develop educational test items, including AI-generated English passages. It also proposed using AI to review questions and predict their level of difficulty.