South Korea’s controversy: Starting in July, image uploads will be matched and screened by AI in advance; netizens blasted it as “internet martial law,” while the authorities denied it was prior review rather than prior censorship.

Starting July 1, South Korea will expand the technical and administrative measures under the Telecommunications Business Act that aim to prevent the distribution of illegally filmed content—from video to images. About 80 obligated businesses will have to automatically compare any image a user uploads against a government-approved database recognized as containing such illegal filming material. The spark was a March 30 announcement by Ruliweb game community administrator, which rapidly turned into controversy and even drew criticism from those calling it a “state inspection.” The official response was that this is not an advance review.
(Background context: The global wave of “community bans” targeting adolescents is accelerating—will Taiwan follow?)
(Additional context: In August, the EU will issue penalties: with no “self-declared identity” for AI, the maximum fine is 3% of global revenue; Taiwan is also within scope.)

Table of contents

Toggle

• The government insists this is not prior review
• Industry backlash
• Do overseas operators end up getting regulated?

Key takeaways

• Starting July 1, about 80 platforms in South Korea that upload images must first compare against a database of illegally filmed content; the thresholds are 1 billion won in revenue from the previous year or 100k daily users
• The government provides a 6-month public-awareness period until December 31; no administrative penalties will be imposed during that time
• Google, Meta, and X are all on the obligated list; the heaviest February 2025 fine of 15 million won was paid by X

Starting July 1, South Korea expands the filtering scope for illegally filmed content from video to images. Under Article 22-5 of the Telecommunications Business Act, businesses with revenue of 1 billion won or more in the prior year, or with 100k or more daily users, must automatically compare—before allowing—each image a user uploads with the government’s digital crime database, to ensure it is not already recognized as illegal filmed content.

The obligated businesses identified by the Korea Communications Commission (방송미디어통신위원회; formerly the Broadcasting and Communications Commission) number around 80. Google, Meta, X, Naver, and Kakao are on the list, and large communities such as DCInside and Ruliweb are included as well.

What ignited the debate was not a government announcement, but a community post. On April 30, Ruliweb (루리웹) game community administrator posted an explanation, saying that illegal filmed content hardly ever appears on such sites, yet they are being made to bear the excessive burden for it. After the article spread, Korean public opinion began using the phrase “state inspection” to describe the entire system.

The government’s technical obligations had been discussed for six years; in the end, it was lifted to the spotlight by an announcement from a game forum—which in itself suggests problems with how the policy was communicated.

The government insists this is not prior review

The regulator’s defense is blunt. The official position is that image matching and identification technology “is not a way of directly reviewing content, nor is it prior review.” It is only meant to block already recognized illegally filmed content from circulating again, and the objects being matched are limited to images and videos approved through deliberation by the Broadcasting and Communications Commission.

Technically, this is called “image DNA.” In plain terms, the system does not store the original image; it extracts feature values into a database, then matches new images against that database. Evasion methods such as cropping, rotating, flipping, blurring, changing colors, and adding text must all be recognized. The government’s set thresholds are a detection rate of 95% or higher, and a normal-image false-blocking prevention rate of 95% or higher.

The government also emphasized that this is not a new obligation. When the law was amended in 2020, the provisions already referred to “illegally filmed content and so on,” which originally covered both video and images; it was simply that the technology had not been able to implement it until the national-level development was completed in December 2025, which is why it entered the schedule then.

Industry pushes back

Industry backlash is focused on costs. A community post may include dozens of images—image traffic far exceeds video. That effectively means additional high-performance GPU servers are needed, and the money is absorbed by the businesses themselves. Upload delays and reduced service quality are the inevitable downstream effects.

Another pain point is misclassification. Some people are already worried that fan art drawings or edited photos could be wrongly intercepted. On the other hand, for illegally filmed content not yet in the database, this mechanism originally cannot block it—this is also the source of the “making it for show” criticism.

The government provided a 6-month public-awareness period, with no administrative penalties from July 1 to December 31. They go live first and push the penalty schedule back by half a year—this has been the typical rhythm of recent years when South Korea introduced platform obligations.

Do overseas operators actually get regulated?

The most widely circulated claim online is that this rule applies only to South Korea domestically, and overseas operators are not subject. This is not true.

Article 22-5 of the Telecommunications Business Act does not differentiate by business nationality; it is based on service scale. Overseas server operators with more than 100k South Korea-based users are regulated in the same way. There is also enforcement history: on February 28, 2025, the regulator fined seven businesses for violating the obligation. The largest fine of 15 million won was imposed on overseas operator X, and more than 10 other overseas businesses received correction orders.

The real loophole is elsewhere. Telegram was not included. The official reason was not “overseas companies,” but that it is a private-message service; one-on-one chatting on KakaoTalk is likewise outside scope. The main distribution channel of the N-room incident at the time was private chat spaces—this is the crux of the effectiveness debate.

Separately, on May 1, large amounts of overseas adult websites began returning HTTP 451 error pages to South Korean connections. As reported by Korean media, this is a restriction implemented by Cloudflare at the request of the South Korean government. Instead of blocking individual websites, the government has moved its “blocking the content” hand to cooperation with infrastructure providers—making it harder to bypass than expanding the law’s applicability.

The controversy exploded because three major developments converged in the same month. On July 7, a revised version of the so-called “fake news eradication law,” the amended Information and Communications Network Act, took effect. Platforms with more than 1 million daily users must publish semiannual transparency reports; those with more than 100k subscribers who intentionally spread fake information to profit through posting can face damages up to five times, with repeat distribution punishable by 1 billion won. On July 21, the amended AI Basic Act took effect again, requiring deepfake content to add markings distinguishable to human eyes.

The People Power Party called the July 7 law the “online gag law.” The Democratic Party rebutted that it is a defense wall to block malicious fake news, and the regulator stressed that the penalties fall on the poster, not the platforms. Korean netizens’ anger may only be starting.

Frequently asked questions

What is South Korea’s image filtering requirement starting in July?

Starting July 1, 2026, about 80 Korean obligated businesses must automatically compare a government database of illegally filmed content when users upload images, to confirm that the content has not already been identified before allowing it through. The public-awareness period runs until December 31, during which no administrative penalties will be imposed.

Does this requirement also apply to overseas platforms like Google and Meta?

Yes. The Telecommunications Business Act does not differentiate by business nationality; it only looks at service scale. Google, Meta, and X are all on the list of about 80 obligated businesses. In February 2025, South Korea fined seven businesses for violating the obligation; among them, the largest 15 million won penalty was paid by overseas operator X.