South Korea - AI Ethics Standards (2020)

Overview

The Human-Centered Artificial Intelligence Ethics Standards were prepared and announced by the Ministry of Science and ICT (MSIT) with technical assistance from the South Korea Information Society Development Institute (KISDI) in late 2020 to set national, voluntary standards to guide ethical AI development and use across sectors. The Standards prioritise 'Humanity' as the highest value and establish a concise set of principles and requirements to operationalise that value from design through deployment. The announcement and accompanying materials are published on official government portals such as Ministry of Science and ICT — national announcement (South Korea.kr) and referenced in international reviews such as the OECD Digital Government Review of South Korea. The Standards are non-binding; their purpose is to provide a durable, adaptable ethics framework that can be operationalised via sectoral guidelines, checklists, voluntary certification, public procurement preferences and educational programs.

Definitions

Key terms used in the Standards include: 'Humanity' — the supreme value protecting human dignity and welfare; 'AI system' — automated systems employing algorithms or models to perform tasks that would normally require human intelligence; 'developer' — any natural or legal person engaged in design, training, testing or maintenance of AI systems; 'operator/provider' — entities deploying or offering AI services; 'stakeholders' — government, private sector, civil society and end-users; 'high‑impact uses' — uses of AI that materially affect fundamental rights, health, safety, or economic livelihood. The Standards keep definitions deliberately broad to accommodate evolving model architectures and application types while anchoring obligations to lifecycle stages (data, design/training, validation, deployment, monitoring).

Governance and Institutional Framework

The Standards propose a distributed governance architecture: MSIT acts as the national coordinator and convener for multi-stakeholder dialogues; KISDI and other research bodies provide technical support and assessment tools; sector regulators (health, finance, transport, communications) interpret and operationalise the principles for domain-specific risks; and the Personal Information Protection Commission (PIPC) oversees personal data obligations under existing privacy law. The Standards recommend establishing inter-agency coordination mechanisms, voluntary standardisation and certification pathways, and public procurement levers to incentivise adoption. The government signalled plans for outreach and capacity building through training, checklists for developers and deployers, and the formation of expert committees — details and follow-up materials were to be published on MSIT and KISDI channels such as KISDI AI Ethics portal. This institutional orientation emphasises soft governance: guidance, incentives and public dialogue rather than immediate statutory obligations.

Key Focus Areas

The Standards identify ten lifecycle requirements as focal points for policy and practice: (1) Guarantee of human rights — preventing violations of dignity and equality; (2) Privacy protection — data minimisation, lawful processing and individual controls; (3) Respect for diversity — inclusive datasets and non‑discrimination by design; (4) Non‑maleficence — minimising foreseeable harms and dual‑use risks; (5) Publicness — recognising public interest and accountability in public-sector AI; (6) Solidarity — protecting vulnerable groups and enabling social inclusion; (7) Data governance — provenance, quality, lineage and stewardship; (8) Responsibility — assignment of roles and human oversight; (9) Safety and robustness — technical validation, resilience and fail-safe measures; (10) Transparency — documentation, explainability and user information. These focus areas span technical, organisational and societal controls and are intended to be instantiated through developer checklists, impact assessments, validation/testing protocols and public communication strategies. The Standards also emphasise continuous monitoring and iterative improvement to ensure the principles remain relevant as technology evolves.

Implementation Framework

Implementation is primarily voluntary and operationalised through: (A) sectoral guidance developed by line ministries and regulators; (B) developer/operator checklists and voluntary self-assessment tools; (C) pilot conformity assessment and certification programs; (D) capacity-building programs for public servants and industry; and (E) public engagement and transparency measures. MSIT signalled that the Standards would be accompanied by concrete instruments — e.g., subject‑specific checklists and educational material — to translate high‑level principles into operational practices for entities of varying size. Where personal data is processed, obligations under PIPC-regulated frameworks remain binding; the Standards point to alignment with existing privacy law rather than supplanting it. The Standards also encourage industry adoption by integrating ethical criteria into voluntary procurement preferences and innovation sandboxes to test compliance approaches before potential future regulation.

Monitoring and Evaluation

Because the Standards are non‑binding, monitoring relies on voluntary reporting, public registries of best practices, pilot audits by research institutions, and the promotion of third‑party voluntary certification. MSIT and partner research agencies were expected to track uptake through surveys, case studies, and public consultations. The Standards recommend periodic review cycles to update guidance in response to technological change and societal feedback; these review cycles were intended to be coordinated with other national initiatives such as the Data and Digital Rights efforts and later sectoral guidance on generative AI and model safety. The OECD review and subsequent government publications track the Standards’ role as a foundation for evolving AI governance in South Korea.

Penalties, Liability, and Appeals

By design, the Standards do not create new statutory penalties — enforcement is soft and reputational: transparency reporting, public procurement considerations, and industry self-regulation. However, the Standards explicitly note that existing legal regimes (e.g., tort liability, consumer protection, and the Personal Information Protection Act/PIPC decisions) remain applicable for wrongful conduct or data misuse. For high‑impact harms, affected parties retain access to civil remedies under general law; administrative enforcement may arise when sectoral regulators translate the Standards into binding requirements. The Standards therefore function as an ethical benchmark that complements, but does not replace, statutory liability frameworks.

Relationship to Other Instruments

The Standards were presented as part of a broader national AI strategy that includes the 2019 National AI Strategy and later instruments such as the Digital Bill of Rights and sectoral guidance. They are explicitly linked to South Korea’s obligations and engagements in international fora (OECD principles, GPAI, UNESCO discussions). The Standards are also intended to coexist with privacy and data protection law under the Personal Information Protection Act; sector regulators are encouraged to interpret the Standards in ways consistent with existing legal duties. Internationally, the Standards are referenced in comparative analyses and OECD reviews as South South Korea’s foundational, voluntary ethics instrument that feeds into later, more prescriptive measures.

International Alignment

South Korea framed its Standards to align with major international instruments, including the OECD AI Principles and related UNESCO and GPAI initiatives. The emphasis on human‑centered values mirrors EU and OECD approaches to trustworthy AI, enabling South Korea to engage in international standardisation efforts and cross‑border cooperation. The MSIT material and subsequent OECD commentary highlight South Korea’s intent to ensure interoperability and mutual reinforcement between domestic guidance and international norms; this approach facilitates participation in global certification conversations and multi‑jurisdictional regulatory dialogues.

Implementation Timeline

Sources and References