AI Regulation Tracker / Regulation in force
South Korea Now Requires AI-Generated Content To Be Labeled, With a Notice for Realistic Deepfakes
Under South Korea's AI Basic Act, any operator offering generative-AI products or services must tell users they are dealing with AI and mark AI-generated outputs. Synthetic media hard to tell from real needs a clear, conspicuous notice.
South Korea's AI Basic Act took effect on January 22, 2026, and with it a duty that reshapes how companies present machine-made content to Korean users. Any operator offering generative-AI products or services must now do two things: tell users in advance that they are interacting with AI, and label the outputs that AI generates. For synthetic audio, image, or video that a person would struggle to tell apart from the real thing, the law requires a clear and conspicuous notice.
The obligation sits in the Act's generative-AI transparency provision, reported as Article 31. Korea's legal portal and several practitioner sources place it there, though the clean statute page was not machine-readable at the time of writing, so it is worth treating the exact section as attributed rather than verbatim. What matters operationally is settled: the transparency duty is live, and it is not a recommendation.
Two ways to mark, one condition attached
The law accepts two forms of marking. The first is human-perceptible: a visible watermark or on-screen label that a user can see for themselves. The second is machine-readable: metadata embedded in the file that software can detect even when a person cannot. Either route satisfies the labeling requirement on its own.
There is one condition worth reading closely. If an operator relies only on the invisible, machine-readable route, it must also surface a "generated by generative AI" notice to the user at least once, through text or voice. In other words, hidden metadata alone is not enough if nothing tells the actual person in front of the screen that what they are seeing or hearing was made by AI. The point of the rule is that a human, not just a detection tool, should be able to know. For product teams that means the invisible route cannot be treated as a quiet compliance shortcut; it comes bundled with a visible or audible disclosure obligation the moment it is used alone.
A higher notice standard for realistic synthetic media
The Act treats convincing deepfakes as their own category. Where generative audio, image, or video is hard to distinguish from real content, the notice must be clear and conspicuous, a higher standard than a quiet metadata tag. This is the provision aimed at synthetic media that could mislead a viewer about what is real, and it applies across contexts rather than to any single sector. A convincing AI voice clone in an ad, a face-swapped video, or a photorealistic generated image all fall within reach when they could pass for genuine, so a brand cannot assume that entertainment or marketing framing exempts it.
What it does not do
The transparency duty is a labeling and notice obligation, not a ban. It does not prohibit generative AI, nor does it require operators to prove their content is harmless before publishing. It governs disclosure: users must be told they are dealing with AI and shown when outputs are AI-generated. Separate provisions of Korea's AI framework, and separate sector laws, carry the heavier obligations around high-impact uses; the transparency rule is the horizontal baseline that touches nearly every generative product.
On penalties, Korean-language press reports an administrative fine for failure to label, with escalation for repeat violations. The amount is unconfirmed, so treat it as reported, not established. The compliance driver is the duty itself, not the disputed number attached to it.
The cross-border angle for US readers
For a US company, this is the closest analog yet to Article 50 of the EU AI Act, which likewise requires disclosure of AI interaction and marking of synthetic content. The United States has no federal equivalent. A US brand that localizes marketing, produces AI-generated imagery, or runs a consumer AI product for Korean audiences inherits a labeling duty it would not face at home. Teams already building for the EU transparency rule will find the Korean requirement familiar in shape, though it carries its own "notice at least once" mechanic when only metadata marking is used. The practical takeaway is that Korea, like the EU, now expects the person on the receiving end to be told, and building that disclosure into the product once is far cheaper than retrofitting it market by market.
Frequently Asked Questions
What exactly changed under Korea's AI Basic Act for generative content?
As of January 22, 2026, operators offering generative-AI products or services must give users advance notice they are interacting with AI and must label AI-generated outputs. Realistic synthetic audio, image, or video requires a clear, conspicuous notice. The duty sits in the Act's transparency provision, reported as Article 31.
Who has to comply?
Any operator offering generative-AI products or services to users in Korea, including foreign providers whose outputs reach Korean audiences. That covers US brands localizing content into Korea, consumer AI apps, chatbots, and image, video, or voice generators.
Is a hidden metadata tag enough to satisfy the rule?
Not on its own. Machine-readable metadata is an accepted form of marking, but if an operator uses only that invisible route, it must also surface a "generated by generative AI" notice to the user at least once through text or voice. A visible watermark satisfies the requirement without the added notice.
What is the penalty for failing to label?
Korean-language reporting describes an administrative fine for failure to notice or label, with an escalation for repeat violations. That figure comes from a single source and has not been independently confirmed, so it should be treated as reported rather than established.
Sponsored Training
Browse the full AI Regulation News tracker
Informational analysis for working professionals, not legal advice. Confirm how any rule applies to your situation with qualified counsel.