Multiple bills moving through the National Assembly would impose watermarking, labeling, and penalty rules on generative artificial intelligence outputs, according to an October 17 report in Asia Business Daily that cataloged competing proposals from Reps. Kim Daesik, Kim Wooyoung, Ahn Cheolsoo, Jo Incheol, and Park Chunggwon. The flurry arrives months after Korea’s AI Basic Act took effect in August 2026, leaving lawmakers to fight over how much duty falls on creators versus platforms.
Supporters argue visible labels and embedded metadata reduce harm from synthetic video in elections and finance; critics warn overlapping regimes could chill lawful editing tools and burden small publishers who lack enterprise detection stacks.
Creator duties versus platform duties
Kim Daesik’s draft emphasizes upstream obligations: anyone publishing AI-altered video for news or advertising must disclose generation methods and retain logs. Kim Wooyoung’s version shifts more burden to hosting services, requiring portals to detect unlabeled synthetic media above traffic thresholds and to downrank repeat offenders.
Ahn Cheolsoo’s bill blends both, with safe harbors for platforms that implement government-certified watermark scanners. Jo Incheol focuses on deepfake impersonation, mandating biometric consent labels when a synthetic voice or likeness targets a recognizable person. Park Chunggwon’s proposal adds criminal penalties for malicious political deepfakes within 90 days of an election.
How the AI Basic Act sets the floor
The August 2026 AI Basic Act already requires high-impact AI providers to document training data and perform risk assessments. It does not, by itself, mandate consumer-facing watermarks on every short-form clip. Assembly bills would layer sector-specific labeling on top, similar to how broadcasting law added captioning rules after baseline telecom statutes.
Legal aides on the Science, ICT, Broadcasting, and Communications Committee said harmonization hearings are likely before any markup, because enterprises cannot maintain three different metadata schemas for one upload.
Technical standards fights
Watermark proponents want interoperable headers readable by major social apps; open-source advocates fear proprietary locks. Korea Internet & Security Agency researchers have piloted frequency-domain marks for video, but lawmakers have not chosen a single standard. Platform lobbyists asked for grace periods so mobile apps can ship decoder updates.
Newsrooms told staff unions they support transparency labels but need clarity on whether color grading or audio cleanup triggers disclosure. Sports and entertainment editors worry fan-edit culture could be swept into punitive tiers if thresholds are vague.
Enforcement and penalties
Park’s election-season criminal clause is the sharpest edge, with fines and possible imprisonment for distributors who knowingly omit labels on synthetic candidate footage. Civil liability bills elsewhere in the packet would let victims sue for likeness misuse without proving monetary loss, lowering barriers for public figures.
The Korea Communications Commission would gain audit powers under several drafts, overlapping with Personal Information Protection Commission roles on biometric data. Coordination memos among agencies are circulating, but no single enforcement map is public yet.
What businesses should track
Media companies should inventory which production tools already export Content Credentials or C2PA metadata, because Assembly staffers cite those frameworks in committee slides even when not named in statute. Ad agencies must watch whether synthetic spokesperson spots require pre-clearance labels before TV broadcast.
For platforms, the practical question is whether Korea mandates server-side detection or accepts client-side creator attestations. The answer will shape cloud costs for short-video apps serving Korean teens—a demographic lawmakers mention often in hearing transcripts. None of the bills is final law yet, but the October 17 docket shows labeling is now a partisan competition, not a single-government decree.
Startup founders told Asia Business Daily that ambiguous “synthetic media” definitions could sweep AI-assisted B-roll and localization dubbing into the same bucket as malicious deepfakes unless committees carve out newsroom and education exemptions in markup.