South Korea AI Basic Act 32: Safety measures and reporting for AI trained above the computing threshold

What does 32 require?

An AI business operator whose AI system was trained with cumulative computing at or above the level set by Presidential Decree implements two sets of safety measures: identification, assessment and mitigation of risks across the whole AI lifecycle, and a risk management system that monitors and responds to AI safety incidents. The operator submits the results of these measures to the Minister, following the detailed guidelines the Minister publishes for the measures and the submission.