Korea AI Basic Act Article 31: labeling an exported AI video

For AI video that is downloaded or shared outside the service where it was made, MSIT's transparency guidelines call for a label people can recognize, such as a visible or audible identifier, or a machine-readable marker added after a text or voice notice. On Sume, the visible route is a text overlay burned onto the MP4 with authored cues.

This is a description of the ministry's press release and Sume's docs, not legal advice. Sources were read 2026-09-30.

What do the guidelines say about exported content?

MSIT unveiled the Guidelines on Ensuring AI Transparency on January 22, for Article 31 of the AI Basic Act. They split content in two: output shown only inside a service environment (screens, UIs, apps), where flexible labeling such as a logo is allowed, and output exported by downloading or similar means, where "more explicit labeling is required". For AI-manipulated video that is hard to tell from reality, the labeling must be clearly recognizable.

The release also says a grace period of at least one year applies, and that fact-finding and penalties under the provision are deferred during it. It says the obligation falls on AI business operators who provide AI products and services, not on users who use them as tools; its example is a film company using AI video tools.

How do I burn a visible Korean label onto a clip?

Send the finished clip to POST /v1/video-captions with cues. Passing text, start and end per cue burns authored overlay copy without speech-to-text, so a silent clip works too. Korean copy needs a Hangul-capable style: slam, punch and tiktok-green return 400 for Korean text, while an omitted style resolves to black-outline for Korean. The clip must be a public HTTPS URL.

curl -X POST https://api.sume.com/v1/video-captions \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: ai-label-001" \
  -d '{
    "video_url": "https://example.com/clip.mp4",
    "style": "black-outline",
    "cues": [
      { "text": "AI로 생성된 영상", "start": 0, "end": 5 }
    ]
  }'

Can I put a badge image on the frame instead?

Timeline compose takes one still and one video and returns one MP4 with both on screen, using operation of stack or overlay. Both URLs must already be media in your workspace, so import first with POST /v1/media-imports. See Timeline compose.

What does Sume not do here?

The caption and compose docs describe burned-in pixels, not metadata. If you choose the machine-readable path, that step is outside what these pages cover; see does Sume add C2PA or a watermark. For the comparable EU rule, read Article 50 video labeling.