AI-Made Photos and Videos Now Need a Label — What Korea's AI Basic Act Watermark Rule Actually Changed
Since January 22, 2026, Korea's Framework Act on the Development of Artificial Intelligence and Establishment of Trust ("AI Basic Act") has been in effect -- often described as the world's first comprehensive AI law. The provision ordinary users run into most often is the disclosure requirement for generative AI output: if text, an image, or video was made with AI, the person or business publishing it must make that fact identifiable to the audience. Until now, whether an AI-made image was labeled as such was left entirely to the creator's conscience or a platform's own policy. The biggest change is that there's now a legal floor for disclosure.
What the AI Basic Act newly requires
The label doesn't have to be a literal "watermark"
The word "watermark" makes people picture a stamp in the corner of an image, but the law and its enforcement decree allow a broader range of methods. An on-screen notice reading "This content was created with AI," or a disclosure in the content's description field, both satisfy the requirement. What matters is that the disclosure is something a viewer can actually notice -- a label that's too small or buried somewhere hard to find may not count as meeting the obligation.
Deepfakes face a stricter rule -- Article 31
Separate from the general AI-labeling rule, content that could be mistaken for a real person or event is held to a stricter notice standard. Article 31, Section 3 requires clear disclosure for this kind of deepfake-style content, with synthesized video or images using a real person's face or voice as the primary target. The bar here is higher than simply noting "AI-generated" -- viewers need to be able to clearly recognize that what they're seeing isn't an actual, real situation.
Fines can reach about $22,000 -- but not starting immediately
Violating the AI Basic Act can bring a corrective order or a fine of up to roughly 30 million won (about $22,000). Enforcement didn't begin the moment the law took effect, though. To reduce confusion on the ground, regulators built in a grace period of at least one year, and actual fines aren't expected to start until 2027 at the earliest. In other words, the obligation itself is already in force, but violating it right now doesn't lead straight to punishment -- it's a transition period. Once the grace period ends, though, any content that was never labeled may need to be fixed all at once, so it's more realistic to treat this window as time to prepare rather than time to ignore the rule.
If you spot AI content with no disclosure
If you come across a photo or video that looks AI-made but carries no disclosure at all, the first thing to determine is whether the content could be mistaken for a real person -- that is, whether it's deepfake-style. If someone's face was synthesized in a way that could enable defamation or fraud, this goes beyond a simple labeling violation and can trigger other laws (defamation, copyright, the Act on Promotion of Information and Communications Network Utilization), which is worth pairing with a report through the platform or a cybercrime report to police. If the issue is simply that the disclosure is hard to notice, though, since we're still in the grace period, asking the platform or creator to make the label clearer is the more appropriate first step rather than reporting it right away.
If your company uses AI-made marketing images
This obligation applies more directly to businesses that provide AI services or produce and distribute content than to individual users. If your company's social accounts or ads use images or video made with generative AI, it's safer to start building a disclosure habit now, even during the grace period. That's an easier path than scrambling to fix an entire content library once the grace period ends and enforcement begins -- building disclosure into the image-generation step itself from the start costs far less than retrofitting it later.
Does this apply to personal social media posts too?
The text of the law defines the parties who distribute AI-generated content fairly broadly, and the general reading is that an ordinary individual's everyday social media posts aren't likely to face immediate enforcement. That said, an influencer with a large following mixing AI images into sponsored content, or anyone effectively distributing AI content for business purposes, could be judged closer to a business under the rule. Clearer boundaries should emerge as the detailed enforcement-decree guidance and real enforcement cases accumulate.
The bottom line
Three things are worth remembering right now. First, the AI Basic Act has been in force since January 22, 2026, and generative AI output carries a disclosure obligation. Second, deepfake-style content faces a stricter notice requirement under Article 31. Third, fines can reach about 30 million won, but because of the grace period, actual enforcement isn't expected until 2027 or later -- this is a transition window. If you're using AI content for business purposes, building a disclosure habit starting now is the safest move.
All content is fact-checked under our editorial standards.