South Korea: AI content labelling rules

The rules on labelling AI-generated content in South Korea, what each asks of a file, and what one Kennmarq.com call adds.

In short

Images: Covered
Video: Covered
Audio: Partly; deep fakes December 2026
Text: Planned, March 2027
3D: Planned, June 2027
Rules: 1 general, 1 for some sectors only.
In the API: markets=kr

Updated 4 November 2026; sources checked 4 November 2026. Sources are government sites. This page explains what Kennmarq.com does; it isn't legal advice.

General rules

AI Basic Act, Article 31: In force since 22 January 2026. Output marked as AI-made (machine-readable if users are told once); deep fakes clearly marked.

Rules for specific sectors

Political ads and campaigns

Public Official Election Act, Art. 82-8: In force since 29 January 2024. Banned 90 days before an election; otherwise a fixed Korean notice, boxed on screen and spoken for sound.

In the API

Send markets=kr on each call to POST /v1/mark: a call that lacks what South Korea's rules need is refused before anything is counted, and each mark gets a receipt of what it carries for each law.

What a mark for South Korea carries, by kind of file
File	Made with AI	A deep fake (`deepfake=true`)
Picture	Watermark and Content Credentials	Watermark, Content Credentials and the visible label
Clip	Watermark and Content Credentials	Watermark, Content Credentials and the visible label
Song	Watermark and Content Credentials	Refused for now: South Korea's deep fakes in sound need a notice people can hear, which we don't add yet.

Mark a picture for South Korea

curl https://api.kennmarq.com/v1/mark \
  -H "Authorization: Bearer $KENNMARQ_API_KEY" \
  -F file=@image.png \
  -F source=created \
  -F markets=kr \
  -o image.marked.png

An AI coding agent gets the same answer from the MCP tool get_requirements (MCP server).

South Korea on the map, and every country's rules side by side.