South Korea Uses AI to Flag 96,210 Child Sexual Abuse Materials as Removal Gaps Persist

10 Min Read
South Korea Uses AI to Flag 96,210 Child Sexual Abuse Materials as Removal Gaps Persist

Faster detection meets the challenge of getting content removed

South Korea's National Center for Digital Sexual Crime Response identified 96,210 items of child sexual abuse material between April and September using an artificial intelligence system that helps staff find suspected content and request its removal. The Ministry of Gender Equality and Family said daily detections reached 202.1 per monitoring worker, compared with 20.8 before the system's introduction, an increase of about 9.7 times.

Contents
  1. Faster detection meets the challenge of getting content removed
  2. How the national system works
  3. What the increase means in numbers
  4. Young people account for most people receiving help
  5. Seoul offers a separate example of faster reporting
  6. Why a removal request may not end the abuse
  7. The next target is grooming in conversations
  8. People remain responsible for support
  9. How the response has developed
  10. Key Points

The system, introduced in April, searches social media, online communities and random chat applications around the clock. Workers review its findings before removal requests go to platforms. Tasks that once required employees to search websites and assemble reports manually can now be processed much faster.

The figures show a sharp increase in the center's ability to find and report material. Getting websites to act is a separate challenge: the center said 28.5 percent of its removal requests went unfulfilled in 2025. Among websites whose server locations it could identify, at least 95.5 percent were hosted overseas.

For victims, the practical stakes extend beyond the first removal request. Images can be copied and uploaded again, requiring repeated searches and continuing support. The national center combines monitoring and removal assistance with help involving legal action.

South Korea's experience also brings together two uses of the same technology. Artificial intelligence can create fabricated sexual images, commonly called deepfakes, while detection systems can help staff identify such material and respond more quickly. Human decisions remain part of the process, both when reviewing suspected abuse and when supporting the people affected.

Advertisement

How the national system works

The national center's system collects posts using about 220 keywords drawn from its monitoring work. It examines text and images together, then presents staff with an assessment and the reasons behind it. A worker makes the final reporting decision.

Once a worker approves a report, the system prepares information required by the relevant platform or authority. Processing a removal request now takes about a minute or less. That is the time needed to prepare and process the request; the website must then respond and remove the content.

Images are blurred by default, and employees reveal them when necessary to reach a decision. This reduces repeated exposure to disturbing material while preserving the review step. The center also intends to improve detection accuracy by training the system with its database.

Kim Mi-soon, director of the National Center for Digital Sexual Crime Response, described how automation now supports the work:

From detecting whether material has been circulated to submitting a takedown request, the system handles the process mostly automatically, without requiring human intervention at every step.

The center has installed data collectors on 14 platforms, including Google, X, Bluesky and Telegram. These collectors automatically gather material for assessment, allowing employees to spend less time searching manually and more time on review and victim assistance.

What the increase means in numbers

The center processed 45,438 cases in all of 2025. Its April to September total of 96,210 in the following year was about 2.1 times that annual figure, achieved within six months. Daily output per monitoring worker rose from 20.8 to 202.1, a gain of 181.3 cases.

The center supported 10,637 victims in 2025, compared with 4,973 in 2020. That is an increase of 5,664 people, or about 114 percent, over five years. Victim totals count people receiving assistance, while processing totals describe the center's work on material and removal requests. One person's case may require repeated action as content circulates.

Advertisement

The ministry plans to expand monitoring to 139 platforms in 2027. Its announcement put current coverage at 74 platforms, although a current total of 79 has also been cited. The center separately describes its expansion of automated collection from 14 platforms to 139. The published figures leave the precise relationship between broader monitoring coverage and installed collectors unresolved.

Using the ministry's figure of 74, reaching 139 would add 65 platforms, an increase of about 88 percent. Wider coverage would bring more places where abusive material may appear into the monitoring process, while also increasing the material employees must assess.

Young people account for most people receiving help

Teenagers and people in their twenties made up 77.6 percent of those supported by the national center in 2025. Teenagers accounted for 28.5 percent and people in their twenties for 49.1 percent. The center assisted 3,052 people under 20, including 20 girls younger than 10.

The center also identified a pattern in which perpetrators approach potential victims through chat applications, quickly build trust and move toward exploitation. A child may be unaware that abusive material is circulating or may be too frightened to seek help. The center therefore searches proactively for child sexual exploitation material, including when no victim has submitted a request.

Police figures provide a separate view of the scale of online sexual abuse. Between November 2024 and October 2025, South Korean police detained 3,557 suspects in online sexual abuse cases. Teenagers accounted for nearly half of the suspects, and offences involving deepfakes represented 35.2 percent of cases.

Deepfakes are images, video or audio altered or generated with AI to make a person appear to do or say something they did not. In sexual abuse cases, fabricated material can place a person's face into sexual content. Kim said the center's system can identify deepfakes that are difficult to recognize by sight.

Seoul offers a separate example of faster reporting

Seoul Metropolitan Government introduced its own AI monitoring system in 2023 and announced automated reporting functions in 2025. In its May 23 announcement, the city said detection and identification that previously took two and a half to three hours could be completed in six minutes.

Advertisement

The city system monitors continuously, generates reports and drafts emails asking hosting websites to remove material. A Removal Support Officer reviews each draft before it is sent. Emails can be prepared in seven languages, including English, Chinese, Japanese and Russian, to reach websites operating abroad.

Seoul also described a narrower processing measure: entering keywords and generating results took three minutes, compared with an average of two hours for officers working manually. The city calculated that reduction at 97.5 percent. The six minute and three minute figures describe different tasks within the process.

Removal cases handled by Seoul's AI system rose from 2,509 in 2022 to 14,256, an increase of about 468 percent. The Seoul Digital Safety Support Center assisted 3,650 individuals over three years and recorded 64,677 support cases, including investigative and legal assistance, psychological counseling and medical care.

Children and teenagers accounted for 50 support cases, or 16.2 percent, in 2022. By 2024, the city recorded 624, or 22.1 percent. That represents roughly 12.5 times as many cases and an increase of 5.9 percentage points in their share. Seoul's municipal service and the national center operate separate systems, so their activity figures belong to different programs.

Why a removal request may not end the abuse

The national center's 28.5 percent rate of unfulfilled requests in 2025 means more than one request in four did not produce the requested removal. Kim also said more than a quarter of websites contacted did not respond at all. Some illegal sites try to evade blocking by changing access routes.

Overseas hosting adds another layer of work. The center cooperates with police and investigative bodies, as well as support organizations abroad and embassies. Sending a request in the right language can help a platform act, while enforcement against an uncooperative operator requires cooperation beyond the center itself.

The Ministry of Gender Equality and Family, police and the Korea Communications Standards Commission have formed a joint victim support group focused on interrupting the circulation of abusive material. Measures described include pursuing deletion at the server level or shutting down illegal websites.

Kim has stressed the need for continuing assistance because content may be copied and distributed again after removal. For a victim, support can therefore involve searching for fresh uploads, making further requests and obtaining legal or psychological help over an extended period.

Advertisement

The next target is grooming in conversations

The national center is developing technology to detect online grooming by examining how conversations unfold. Grooming is a process in which an adult or other perpetrator builds a young person's trust to draw them toward sexual exploitation.

Keyword searches have limits here because an individual message may contain ordinary language. A comment about someone's appearance or a question about their age can become part of a more concerning exchange when followed by pressure to move into private contact or sexual activity. The planned technology would assess the sequence and context rather than rely only on isolated words.

The center is also developing an AI chatbot to give vulnerable young people information about support services. Both projects extend its work beyond finding material already online: one aims to recognize approaches that may lead to exploitation, while the other aims to make assistance easier to find.

No deployment date was given for the conversation analysis tool or chatbot. The dated expansion target is 2027, when the center plans to bring its AI coverage to 139 platforms.

People remain responsible for support

The national center began operating in 2018 and was later renamed following changes to legislation governing sexual violence prevention and victim protection. It operates within the Women's Human Rights Institute of Korea, with four teams and an authorized staffing complement of 43.

Automation changes how those employees spend their time. Searching continuously, grouping findings and preparing request forms can be handled partly by software. Deciding what action to take, helping someone pursue legal remedies and responding to distress still require staff.

Park Kwang-soo, head of the Women's Human Rights Institute of Korea, described the place of people in the service:

Whatever AI or technology we use, victim support must begin and end with people," Park said. "And people are needed at countless points in between.

Blurred previews, automated paperwork and faster collection serve that approach by reducing repetitive work and exposure to harmful images. The center's expansion will depend on both finding more material and maintaining the human review and support needed to act on it.

Advertisement

How the response has developed

South Korea's national and municipal programs have developed through several stages. Their timelines show how victim assistance has expanded alongside automated searches and reporting.

  • 2018: The national digital sexual crime victim support center began operating.
  • 2023: Seoul Metropolitan Government introduced its AI monitoring system.
  • November 2024 to October 2025: Police detained 3,557 suspects in online sexual abuse cases.
  • May 23, 2025: Seoul announced automated reporting functions, with officer review before removal emails are sent.
  • 2025: The national center supported 10,637 victims and processed 45,438 cases.
  • April to September 2026: The national center processed 96,210 cases following the introduction of its AI system in April.
  • 2027: The national center plans to expand AI coverage to 139 platforms.

The next scheduled expansion concerns platform coverage. Its practical benefit will depend on how quickly suspected material reaches human reviewers and how effectively platforms and authorities respond to the resulting requests.

Key Points

  • The national center identified 96,210 items of child sexual abuse material between April and September.
  • Daily detections per monitoring worker rose from 20.8 to 202.1, about 9.7 times the previous level.
  • Workers review suspected material before removal requests are submitted.
  • In 2025, 28.5 percent of the national center's removal requests went unfulfilled.
  • Teenagers and people in their twenties accounted for 77.6 percent of victims receiving national support.
  • Seoul operates a separate AI system that drafts removal emails in seven languages for officer review.
  • The national center plans coverage of 139 platforms in 2027 and is developing tools to identify grooming and share support information.
Share This Article

You May also Like