Korea AI Basic Act grace period: when do 30 million won fines start?

Korea's AI Basic Act took effect in late January 2026, and the Ministry of Science and ICT announced a grace period of at least one year before it imposes fines, so fines for labeling failures should not start before roughly January 2027. The maximum administrative fine in the sources I read is 30 million won.

The sources are Shin & Kim's enforcement update and Korea JoongAng Daily's explainer, both read on 2026-10-02. They differ on the effective day by one (22 or 23 January), so I say late January. I did not read the statute or the enforcement decree text.

What does the grace period cover?

Shin & Kim says the ministry set a grace period of at least one year beginning on 22 January 2026. During it, enforcement focuses on guidance rather than immediate penalties, except for serious cases involving loss of life or human rights violations. JoongAng Daily adds that investigations are expected to be suspended during the period.

'At least' matters. The ministry has not, in these sources, fixed an end date, so do not plan around a specific day in January 2027.

What obligations is the grace period postponing?

Per Shin & Kim, generative AI service providers must indicate when content is machine-generated. For ordinary outputs, either a human-recognizable label or a machine-readable method such as a watermark may be used. For deepfakes, only labeling that users can clearly recognize is permitted. Providers must also give users advance notice that generative AI is in use.

The grace period does not change those duties; it changes when penalties begin. Teams that wait until fines start will have no time to retrofit a delivery pipeline.

How can a Sume user prepare now?

Preparation is mostly a pipeline question. Decide where the visible label lives, and put it in the pixels so it survives every re-encode. Video captions burns authored cues with text and start and end times, which suits a short notice. Keep each job id and delivered artifact from Jobs and results as a record of which clip was AI-generated and when.

What Sume does not do: it does not decide whether your use counts as a regulated service, and its docs do not describe a watermark or embedded label on outputs. Sume's docs describe what each tool returns and say nothing about a watermark or embedded provenance record on outputs, so treat the output as unmarked until you have checked the delivered file yourself.

Why does the date matter for a video team?

A grace period is a planning window, and for a team that renders AI video it is the window to build the label step once rather than patch it per campaign. Changes to a delivery pipeline touch exports, review and storage, so they take longer than a single clip edit suggests.

The cost structure is lopsided. A visible notice costs seconds of screen time and one extra render. A corrective order, which Shin & Kim names for labeling violations, costs a rework under pressure, and the fine for advance-notice violations is capped at 30 million won in the sources.

JoongAng Daily also reports that only 2 percent of 101 surveyed Korean AI startups had substantive compliance systems before the law took effect. If that figure is right, most competitors are in the same position, so being ready early is a modest advantage rather than a requirement you are late on.

What should you not assume?

Use this as a working list and adjust it to your own pipeline and counsel's advice.

• That the grace period means no duty: it only defers penalties, and serious cases are carved out.
• That 30 million won is the ceiling for every violation: the sources tie it to specific ones.
• That an end date exists: the ministry said at least a year.
• That a source summary replaces the decree: read the decree text with counsel.