South Korea AI law enforces deepfake labels and frontier safety rules

South Korea just moved from talking about AI regulation to actually enforcing it, and that’s a meaningful inflection point in the race to AGI. The AI Basic Act doesn’t just talk about principles; it binds high‑impact and frontier‑scale systems to concrete duties around transparency, watermarking, and safety, with explicit attention to deepfakes and AI‑generated misinformation. That makes Korea the first country where operating a powerful model without clear labelling and safeguards isn’t just bad optics—it’s illegal.

This matters because the long‑running argument that “regulation is years away” is now demonstrably false. Any frontier lab or platform that wants meaningful access to the Korean market now has to implement watermarking, disclosure flows, and risk governance that regulators can actually inspect. In practice, those compliance systems tend to get reused globally rather than rebuilt for each jurisdiction, so Korea is likely punching above its weight in setting norms for watermarking and high‑risk AI. The law also carves out sensitive domains like nuclear, healthcare and finance for extra scrutiny, foreshadowing how other countries may handle AGI‑class systems.

The competitive angle is subtle: the government is explicitly pitching this as both a safety and industrial policy tool, aiming to make Korea a top‑three AI power. Expect local champions in chips, cloud and models to lean into “K‑compliant” offerings, while global players like OpenAI and Google face another layer of regulatory complexity when deploying in Asia.