Overview
- The AI Basic Act took effect on Jan. 22, requiring human oversight for high-impact uses and clear disclosure or watermarking when AI output could be mistaken for reality.
- Enforcement centers on compliance with corrective orders, a minimum one-year grace period, no criminal penalties, and fines capped at 30 million won for ignored orders.
- Korea defines duties for high-performance systems using technical thresholds such as cumulative training compute, which officials say no current models meet.
- The law establishes national governance through the Presidential Council on National Artificial Intelligence Strategy and a legal basis for an AI Safety Institute.
- Startups warn of vague requirements and potential burdens, as the government signals possible grace-period extensions and requires large foreign providers to appoint local representatives, a bar currently met by OpenAI and Google.