South Korea's AI Basic Act, a 43-article law the government bills as the first of its kind enforced nationwide, came into force on January 29. Officials pitch it as building trust in AI while supporting the goal of becoming one of the top three AI powers.
Companies offering AI services must label what their systems produce. Citing The Guardian, PetaPixel reports that obviously artificial work such as cartoons or stylised art needs invisible watermarks, while realistic deepfakes need visible labels. Fines reach 30 million won, about $20,400, after a grace period of at least a year.
The law also covers high-impact systems used in areas such as medical diagnosis, hiring and lending, which need risk assessments and documentation. If a human makes the final decision, a system can fall outside that category. Developers of the most powerful models must file safety reports, though officials concede the threshold is set so high that no model now in use anywhere meets it.
Local startups say the labelling and disclosure rules go too far and may slow them. The government counters that 80 to 90% of the act is about promoting the AI industry rather than restricting it. The EU is phasing in its AI Act through 2027, including disclosure of copyrighted training data, while the White House has moved against state AI laws it considers too strict.
Sources · 1