Korea's AI System Found 96,210 Child Abuse Items in Six Months

Korea's new AI detection system identified 96,210 items of online child sexual abuse material between April and September, a tenfold increase over last year's detection pace, according to the gender equality and family ministry.

The National Center for Digital Sexual Crime Response, which introduced the AI-powered system in April, flagged an average of 202.1 cases per day during the six-month period. Last year, the center's daily average stood at 20.8 cases. The ministry announced the figures on Tuesday.

The gap between those two numbers is the core of the story. Human monitors, working without automated support, located roughly one-tenth of what the AI-assisted system now finds every day. The ministry attributes the increase directly to the round-the-clock monitoring capability the system provides.

Why the numbers jumped so sharply

The system scans online platforms in real time, operating continuously rather than during staffed hours. It monitors social media services and random chat apps, two categories of platform where illegal material has proven difficult to track through manual review.

Before the AI deployment, employees at the center monitored platforms in person. That approach limited both coverage and speed. The ministry said the automated system reduces the workload on those employees, allowing the machine to perform the initial scan and the humans to focus on judgment calls.

The division of labor follows a now-standard pattern in content moderation: the AI system flags material it identifies as child sexual abuse material, and center workers review each flagged item before the center files a removal request with the platform hosting it. No removal request goes out on the AI's judgment alone.

This human-in-the-loop structure (a workflow where automated decisions require human confirmation before action) matters for accuracy. False positives in this domain carry serious consequences for wrongly flagged users, while false negatives leave illegal material online. The ministry has acknowledged both risks and said it plans to improve the system's accuracy by training it on the center's own database of previously identified material.

What changes for the platforms?

The center does not remove material itself. It notifies the platforms, which then carry out the removal. Between April and September, the notification pipeline processed 96,210 items across the platforms under monitoring.

This year the system covered 79 online platforms. Next year, the ministry plans to expand monitoring to 139 platforms, a 76 percent increase in coverage.

That expansion places a growing compliance burden on Korean platforms and on international services operating in the country. Random chat apps, which pair anonymous users for short conversations, have been a particular focus of Korean digital sex crime enforcement for years, because their ephemeral and anonymous design complicates both detection and evidence collection.

Why Korea built this system

The National Center for Digital Sexual Crime Response was established in 2018 to provide comprehensive support to victims of digital sex crimes. Its mandate includes removing illegal online material and taking legal action on victims' behalf.

The center's caseload illustrates the scale of the problem it addresses. Last year it supported 10,637 people. Of those, 3,052 were victims under the age of 20 — nearly 29 percent of the total.

Korea is one of the most highly connected countries in the world, and the ministry framed the AI deployment as part of Seoul's effort to counter digital sex crimes that have evolved alongside that connectivity. The ministry said the adoption of the AI system reflects the government's assessment that manual monitoring can no longer keep pace with the volume and distribution methods of illegal material.

The child sexual abuse material category is one part of a broader digital sex crime problem the center handles, which has included non-consensual filming, forged imagery, and blackmail schemes built on intimate content. The center's victim support role spans removal requests, legal action, and assistance services.

The detection pipeline, step by step

According to the ministry, the process works as follows:

• The AI system monitors the 79 platforms continuously, in real time, day and night.
• The system flags material it identifies as child sexual abuse material.
• Center workers review every flagged item.
• After human confirmation, the center files a removal request with the platform.
• The platform removes the notified material.

The ministry did not disclose the system's error rates, the specific detection techniques it uses, or the names of the 79 platforms currently monitored. It also did not break down the 96,210 items by platform type or by whether the material involved Korean victims.

What comes next year?

Two developments are scheduled for next year. First, the expansion from 79 to 139 monitored platforms. Second, continued training of the system using the center's database, which the ministry expects will improve detection accuracy.

The 96,210 figure covers six months. If the expanded system sustains or improves on the current daily average of 202.1 cases across nearly double the platform coverage, the annual identification count for next year could substantially exceed anything the center recorded in its pre-AI history.

The ministry's announcement positions Korea among the governments now applying AI detection systems to child sexual abuse material at national scale. The results reported Tuesday — a jump from 20.8 to 202.1 daily detections — will serve as an early data point on whether automated monitoring, paired with human review, can meaningfully reduce the availability of such material on mainstream platforms.