South Korea - AI Responsibility Act

Overview

The Artificial Intelligence Responsibility Act (commonly referred to in Korean coverage as the "AI Responsibility Bill" or "인공지능책임법") was introduced as a stand‑alone draft law by Representative Hwang Hee on 2023-02-28. The draft law sought to establish statutory principles for safe and trustworthy artificial intelligence, to allocate duties between the State and AI operators, to define a category of "high‑risk AI", to create enforceable user rights (including explanation and remedies), and to establish institutional mechanisms for dispute resolution and oversight. The bill was framed as a comprehensive legal attempt to promote beneficial AI uses while restraining or mitigating AI applications that may harm life, bodily integrity, property, or fundamental rights. The proposal was part of legislative activity during the 21st National Assembly but did not complete the legislative process; the draft is recorded as ended (lapsed) with the close of that legislative term.

Definitions

The draft bill included working definitions for core terms used throughout the text, including "artificial intelligence system", "developer", "operator", and "user". The most notable definitional element was the proposed legal concept of "high‑risk AI", which targeted AI uses and deployments that pose a significant risk to life, safety, property, or fundamental rights. Definitions were designed to be technology‑neutral, intended to capture both algorithmic decision‑making systems and generative/learning models insofar as they affected or were deployed in regulated domains. The bill aimed to ensure definitional breadth sufficient to encompass emergent AI applications while enabling risk‑based differentiation of regulatory duties and obligations.

Governance and Institutional Framework

The draft established several institutional elements to implement and oversee the proposed regime. It allocated coordination responsibilities to relevant ministries, notably identifying the Ministry of Science and ICT in coordination roles, and contemplated periodic preparation of a national AI basic plan. The bill proposed the creation of a specialized AI dispute‑mediation or adjustment committee/council to adjudicate or mediate AI‑related disputes and to provide a focused mechanism for resolving claims of harm or for facilitating remediation. The bill anticipated interagency coordination for classification of high‑risk uses and for supervisory activity, envisaging that sectoral authorities and central ministries would collaborate on rulemaking, supervision, and guidance to operationalize obligations. The proposal also referenced the need for standards, conformity mechanisms, and potential insurance arrangements as governance tools.

Key Focus Areas

• Definition and scope of "high‑risk AI" — the bill proposed a legal category for AI applications that impose a substantial risk to life, safety, property, or fundamental rights, with examples including biometric identification, critical infrastructure, personnel assessment, emergency services, credit or essential public services, state investigative/immigration uses, and other high‑impact domains.
• State duties and planning — obligations for the State to prepare and periodically update a national AI plan, coordinate policy across ministries, and establish supervisory or advisory bodies and processes.
• Operator duties — mandatory risk assessments, adoption of technical and organizational measures (robustness, security, data governance), maintenance of documentation and logs, and obligations for pre‑deployment assessment and post‑deployment monitoring for high‑risk systems.
• Transparency and user rights — duties to inform affected users, provide meaningful explanations or information about substantive decisions that affect individuals, and ensure access to remedies and dispute resolution.
• Dispute resolution and remediation — establishment of a specialized AI dispute‑mediation/adjustment body to handle AI‑specific complaints, facilitate compensation, and coordinate remedies alongside existing civil liability channels.
• Cybersecurity, data governance and non‑discrimination — measures to address secure development and operation, recordkeeping of training and validation data, and safeguards against discriminatory outcomes.
• Insurance and financial resilience — provisions encouraging or requiring insurance arrangements or other compensation mechanisms for harms arising from high‑risk AI.
• Standards and conformity — encouragement of industry standards, conformity assessment procedures, and certification or documentation regimes to demonstrate compliance and support supervisory oversight.

Implementation Framework

The draft adopted a risk‑based regulatory architecture. High‑risk AI systems were subject to stronger and more prescriptive duties — including pre‑deployment impact assessments, mandatory technical safeguards for robustness and security, stricter transparency requirements, ongoing monitoring obligations, and the maintenance of detailed development and deployment documentation (such as training data provenance, validation and testing procedures, and change logs). The bill required operators to prepare records to enable oversight and post‑incident review and envisioned that regulators would issue technical guidance and rulemaking to operationalize requirements. It also recommended phased implementation tied to risk classification: immediate obligations for clearly high‑risk uses, and phased or staged obligations for other categories to allow time for standards and conformity mechanisms to be developed. The bill anticipated that formal rulemaking and supervisory instruments would follow enactment to define technical standards, impose conformity assessment procedures, and allocate supervisory responsibilities to competent bodies.

Monitoring and Evaluation

Monitoring and evaluation under the proposed regime were to be supported by supervisory recordkeeping, interagency coordination, and the proposed AI dispute‑mediation/adjustment committee. Operators were required to maintain development and deployment documentation and logs (including training data records, validation and testing outcomes, and impact assessments) so that regulators and adjudicative bodies could review systems after incidents or complaints. The bill envisaged ongoing monitoring obligations for high‑risk systems and empowered regulators to require reporting, conduct audits, and issue binding directions where systems posed imminent dangers. The draft also anticipated regular national AI planning cycles (for example multiyear AI basic plans) to assess policy outcomes, update risk classifications, and refine supervisory and technical standards over time.

Penalties, Liability, and Appeals

The draft provided for administrative enforcement mechanisms including corrective orders, mandatory mitigation steps, disclosure duties, and potential administrative sanctions for non‑compliance with substantive obligations. It envisaged facilitation of compensation and remediation for victims via the specialized dispute‑mediation body in addition to ordinary civil liability channels, and allowed regulators to issue binding directions for high‑risk AI systems that posed imminent danger. The bill sought to balance administrative enforcement tools (orders, fines, mitigation requirements) with dispute resolution pathways geared to AI‑specific harms and to enable victims to obtain practical remedies. Specific penalty amounts or precise civil liability reforms were not finalized in the draft text as introduced; enforcement was framed through the combination of administrative powers and access to mediation/adjustment mechanisms.

Relationship to Other Instruments

The draft was explicitly designed to sit alongside Korea's sectoral regulatory framework and existing statutes rather than to wholly displace them. It referenced coordination with sectoral laws governing transport, medical devices, telecommunications, and other safety regimes, and with personal data protection requirements under Korea's Personal Information Protection Act. The bill aimed to fill gaps specific to AI responsibility that sectoral laws or existing privacy and consumer protections might not fully address, proposing coordination mechanisms so that AI‑specific obligations could be harmonized with sectoral supervision and data protection enforcement.

International Alignment

The bill referenced international regulatory developments and suggested Korea align its risk‑based approach with leading jurisdictions. While the draft did not prescribe adherence to any single foreign instrument, it signalled an intent to consider comparative regulatory models and to align principles and risk‑based categorization with international practice to the extent possible. The proposal intended to position Korean regulatory policy in dialogue with global developments in AI governance and to draw on evolving international standards and guidance in the design of conformity assessment and supervisory arrangements.

Implementation Timeline

Sources and References