"Vehicle plate number 2***** KoXX, passed Donghae IC at 6:30 p.m., expected exit Namgangneung IC, average travel time 18 minutes, 193 unpaid cases."
On Feb. 8, a text message pinged the work cellphones of employees in the Toll Management Department at the Gangwon Headquarters of the Korea Expressway Corporation (KEC). It was an alert sent by artificial intelligence (AI) that predicted the route of a "high-amount toll delinquent vehicle" that had entered the expressway. The enforcement team actually caught the vehicle near Namgangneung IC, as flagged by the AI. The driver paid the overdue tolls of 3.51 million won in full on the spot. Without AI, the enforcement squad would have had to wait for hours at chokepoints for a delinquent vehicle to appear, repeating the process.
On the 12th, we visited the Toll Management Department at the Gangwon Headquarters of the Korea Expressway Corporation (KEC) in Wonju, Gangwon. With the Lunar New Year holiday approaching and expressway traffic surging, the AI-based "delinquent vehicle route prediction enforcement system (AI enforcement system)" was operating around the clock.
Lee Juhyeon, 52, deputy head of the Toll Management Department, said the AI enforcement system continues to prove its worth. Lee, who has handled enforcement for seven years, said, "This year alone we caught four vehicles with high unpaid amounts and collected 48 million won."
◇ When a high-amount delinquent car appears, AI sends an 'alert text' to the squad
The AI enforcement system had a thorough grasp of the driving habits of vehicles with high unpaid tolls. It analyzed which expressway entrances and exits they usually used, on which days and at what times they mostly traveled, and their average speed.
When a high-amount delinquent vehicle entered the expressway, the AI enforcement system identified the license plate and sent an alert to the enforcement team. The text message included the hit probability, expected exit, estimated arrival time, and the number and amount of unpaid tolls.
An enforcement car waits near the exit predicted by AI and uses a camera installed on the passenger side to identify the target. Once spotted, they use a lighted baton to guide the vehicle to the nearest IC and collect the overdue amount.
Vehicles that have not paid 3 million won or more in tolls and additional tolls combined are subject to enforcement. Additional tolls are imposed on vehicles that have habitually or repeatedly failed to pay tolls 20 or more times within the past year, or that have not paid despite three notices. Ten times the toll is collected.
◇ Hit rate 25%→90%… "We don't wait anymore"
The Korea Expressway Corporation (KEC) introduced the AI enforcement system in June 2024. Unpaid expressway tolls have surged since Hi-Pass opened. Unpaid amounts jumped more than 50% in three years, from 65.6 billion won in 2022 to 99.4 billion won in 2025. In particular, as of the end of last year there were 4,467 high-amount delinquent vehicles nationwide, with arrears totaling 20.2 billion won.
On the ground, staff said the way they work has completely changed since the AI enforcement system was adopted. In the past, they had only the vehicle number and expected exit information for high-amount delinquents. The enforcement team had to estimate the route and arrival time for each case. It was not uncommon to wait for hours and come up empty-handed.
Kim Changsun, 46, deputy head of the Toll Management Department, said, "We had to check, vehicle by vehicle with habitual nonpayment, which days they traveled regularly and which regions they frequented," and added, "It took considerable time to decide the enforcement point and timing, and field efficiency was not high."
Before AI was introduced, the enforcement hit rate was about 25%—one catch for every four deployments. After using the AI enforcement system, the hit rate has risen to 90%.
That has translated into results. Nationwide on-site enforcement collections have increased each year, from 1.3 billion won in 2023 to 1.8 billion won in 2024 and 2.3 billion won in 2025.
◇ Prediction by AI, enforcement by people… pursuits over 100 km
As the estimated arrival information for high-amount delinquent vehicles became clearer under the AI arrival system, field enforcement methods also became more systematic. For example, when multiple high-amount delinquent cars appear at the same time, the team sets priorities for enforcement based on "prediction probability" and "amount in arrears."
Choi Wonseo, 50, Director of the Toll Management Department, explained, "Vehicles that travel on a regular, repeated schedule are given lower priority because they are likely to be caught again later, and we first enforce against irregular vehicles with high prediction probabilities and large amounts."
Of course, AI can only help with prediction. Enforcement is still the domain of people. Some drivers of high-amount delinquent vehicles notice the enforcement and speed up or change their routes. There was even a case where a chase that started in Wonju, Gangwon, continued more than 100 km to Ansan, Gyeonggi.
Field know-how also fills gaps in the AI enforcement system. In sections where the road splits into two, the AI enforcement system presents a 50% probability that a high-amount delinquent car will head to each exit. In such cases, the enforcement team waits in the section before the fork.
Deputy head Lee Juhyeon said, "The AI enforcement system continues to learn from wrong predictions and is evolving," adding, "We will combine human experience to collect to the end."