A multinational classifies one of its products as a high-impact AI system within the meaning of South Korea's AI Basic Act and is mapping the substantive duties the Act attaches to that classification, as distinct from the broader duties it owes for any generative output. Which two obligations does the Act most directly impose on the operator of a high-impact AI system? Select TWO.
- ACarry out measures to ensure meaningful human oversight and management of the high-impact system's operation and foreseeable risks. Correct
- BNotify and obtain advance clearance from the supervisory authority before each individual high-impact inference is generated for a user.
- CEstablish and operate risk-management and safety measures across the lifecycle of the high-impact AI system to protect users' rights. Correct
- DUndergo a third-party conformity assessment by a notified body before the high-impact system may be placed on the Korean market.
- ERegister the high-impact system in a centralised public EU database before deployment and renew that entry annually thereafter.
Why A is correct: Correct: the Act ties high-impact classification to duties such as putting human oversight and risk-management measures in place for the system's lifecycle.
Why B is wrong: Tempting because high-impact status implies scrutiny, but the Act sets lifecycle safety and oversight duties, not per-inference pre-clearance of every single output.
Why C is correct: Correct: the Act requires operators of high-impact systems to run lifecycle risk-management and safety arrangements aimed at protecting affected users.
Why D is wrong: Tempting by analogy to the EU AI Act's notified-body route, but Korea's Act relies on operator duties rather than mandatory notified-body conformity assessment.
Why E is wrong: Tempting as it echoes a real registration idea, but the EU database is an EU AI Act mechanism and is not a duty imposed by Korea's AI Basic Act.