Check What Developers of Generative AI Services Must Disclose Under Korea's AI Basic Act
This page contains Coupang Partners affiliate links; the operator may earn a commission.
Key Takeaway
Under Korea's AI Basic Act, in force since January 22, 2026, services with generative AI must tell users in advance that AI is used. High-impact AI is covered too, and fines get a grace period of at least a year.

What transparency duties does the AI Basic Act place on developers?
The Policy Briefing article (January 21, 2026) on the Ministry of Science and ICT announcement confirms four duties.
| Item | What it says |
|---|---|
| Advance notice | Products and services built on high-impact AI or generative AI must tell users in advance that AI is used. The enforcement decree accepts a range of methods |
| Labeling outputs | Outputs of generative AI and deepfakes must be labeled so users clearly recognize them, taking factors such as age into account |
| Watermarks | For animation, webtoons, and similar content that is not a deepfake, an invisible digital watermark is also allowed |
| User notice | Users must be told through a notification or the UI that an output was generated by AI |
When you design this, start by deciding where on the screen the AI-use notice goes and how the label will be attached to generated outputs. The detailed labeling standards are set in the enforcement decree and guidelines, so check the original text.
How do you tell whether your service is high-impact AI?
The article explains that it depends on whether the AI is used in a field the law specifies and on how large the risk is. It also says that if a person makes the final decision and can control the system, the AI is excluded from high-impact AI.
If you build features that automate judgments, such as hiring or loan screening, make clear in the design whether a person actually takes part in the final decision. Whether a service qualifies can differ from case to case, so you cannot settle it from the article alone.
What changes during the grace period?
The government has said it will suspend regulation for at least a year, running fines and fact-finding investigations with a focus on guidance in the meantime. Fact-finding is reserved for extremely exceptional cases, such as harm to human life or violations of human rights. The article gives neither the maximum fine nor the date the grace period ends.
So rather than putting it off because of the grace period, it is safer to build the labeling and notice features in early.
Where can you ask questions?
The Ministry of Science and ICT runs an AI Basic Act support desk staffed by experts who took part in drafting the subordinate regulations, and says it keeps consultations confidential and offers anonymous consultation. The March 31, 2026 press release says 552 consultations had been received in the first 10 weeks after enforcement and that a casebook had been published.
That press release page does not show a dedicated contact for the support desk. Check the ministry's notices or guidance pages separately for how to submit a request.
In what order can a development team prepare?
- List whether your service has generative AI features and whether the outputs are shown directly to users.
- Decide where on screens and in API responses you will tell users that AI is used.
- Review technically whether you can attach labels or watermarks to generated outputs.
- If you touch a field the law covers, such as hiring, lending, or healthcare, document the procedure in which a person makes the final decision.
Laws and guidelines can be revised. This article is based on two items checked on 2026-10-08, a Policy Briefing article and a Ministry of Science and ICT press release, so check the enforcement decree and the latest guidelines before you apply any of it.
Frequently Asked Questions
Will fines start right after the grace period ends?
The Policy Briefing article says regulation will be suspended for at least a year, and that fines and fact-finding investigations will focus on guidance during that time. The exact end date and the standards that follow are not given in this material. Check the latest notice from the Ministry of Science and ICT.
How do I know whether my service counts as high-impact AI?
The article says it depends on whether the AI is used in a field set out in the law and on how large the risk is. If a person makes the final decision and can control the system, it is excluded from high-impact AI. For whether your own service qualifies, ask an official channel such as the support desk.
Can I use the support desk anonymously?
The Policy Briefing article says consultation content is kept confidential and anonymous consultation is available. Check the Ministry of Science and ICT's notices for how to submit a request.
Sources & Verification Date
- The AI Basic Act takes effect on the 22nd (Korea Policy Briefing news, 2026-01-21) · Verified
- AI Basic Act support desk logs 552 consultations in the 10 weeks since enforcement (Ministry of Science and ICT press release, 2026-03-31) · Verified
All materials have been verified from official announcements and may have changed since verification.