South Korea has introduced what it calls the most extensive AI regulations in the world, aiming to set a precedent for other nations. However, these new laws have sparked debate among local tech startups, who argue they are overly restrictive, while civil society groups feel they don’t go far enough. The legislation requires companies to label AI-generated content with invisible digital watermarks for clearly artificial creations like cartoons, and visible labels for realistic deepfakes. For “high-impact AI” systems—such as those used in medical diagnoses, hiring, and loan approvals—operators must perform risk assessments and document decision-making processes. If a human makes the final call, the AI may be exempt. Extremely powerful AI models will need safety reports, but the bar is set so high that no current models worldwide meet the criteria.
Back
