South Korea's Interior Ministry has finalised a Public-Sector AI Ethics Standard with six values, a 68-item self-check and human oversight for AI-assisted administrative decisions

Key takeaways

• MOIS says the final standard rests on six core values and a 68-item self-check list.
• Self-assessment has three tiers, by how far AI is used in a service: application, utilisation for decision support, and fusion with autonomous decisions.
• The accountability standard stresses that final responsibility and control stay with people, with human oversight and remedy systems.
• The release states no effective date, penalty or legal basis.

What did MOIS publish, and is it binding?

MOIS announced on 1 October 2026 that it had finalised the standard, whose Korean title is 공공부문 인공지능 윤리기준. The release says the standard was prepared as one of the key tasks under the national agenda of realising the world's best AI democratic government. A draft prepared in November 2025 went through review and comment gathering by related agencies and academic experts before the final text.

The release does not say the standard is binding. It cites no statute, sets no deadline and names no sanction. It describes the document as a practical guideline that officials can use to check and improve their own AI ethics, and says it is "단순히 선언적 가치에 머물지 않고" (not merely a declaration of values). We therefore treat it as non-binding guidance for public bodies, and we state that as our reading, not as MOIS's words. The release also notes that the Ministry of Science and ICT recently enacted AI ethics principles that apply across all sectors, and that MOIS worked on this standard after analysing domestic and foreign AI ethics standards and taking expert advice.

What are the six values and the 68-item self-check list?

The release says the final standard has two goals, promoting administrative innovation through public-sector AI and building public trust, and rests on three perspectives: the public, administration and technology. Its six core values (6대 핵심가치) are publicness, equity, transparency, accountability, safety and privacy protection. The attached annex gives each a one-line description:

• Publicness: public AI is provided for the public interest and to improve quality of life.
• Equity: public AI is provided fairly and without discrimination to all citizens.
• Transparency: the process of adopting and using public AI is disclosed openly.
• Accountability: the party responsible for public AI is clearly set, and the AI is used under institutional standards.
• Safety: public AI is operated in a safe way so that it does not harm citizens.
• Privacy protection: safeguards are put in place so that public AI does not infringe personal information or private life.

From these values the release says MOIS derived 68 detailed check items in a checklist, adapted to the nature of public work. It also introduces a three-tier self-assessment system so that each agency can apply the standard to its level of AI use: tier 1, application (basic use); tier 2, utilisation (decision support); tier 3, fusion (autonomous-decision based). The 68 items themselves were not in the material we opened.

What does the accountability standard say about AI-assisted decisions?

On accountability, the release says the standard focuses on making clear that final responsibility and control rest with people, including when AI supports administrative decisions or is used in direct judgments. It names welfare recipient selection, permits and licensing, and risk prediction as examples of administrative work that directly affects citizens' rights and lives, and says it matters who checks and answers for an AI error in those cases.

On that basis, the release says the standard has agencies clarify who is responsible across the whole AI service, from planning to provision, and set up two things. The first is a human oversight system in which people can review and intervene in AI judgments and decisions. The second is follow-up management and a remedy system for harm. The Korean text reads: "인적 감독 체계와 사후관리‧피해구제 체계를 마련하도록 하였다." The release does not describe what a remedy system includes or how agencies would be checked.

What does it mean for suppliers and administrative-law practitioners, and what to do now?

This section is our analysis. The release addresses public bodies and staff. It does not mention procurement, vendors, audit or contract terms, so any supplier consequence is an inference.

• Suppliers of AI to Korean public bodies can expect customer questions that track the six values and three tiers. Suppliers can map product controls to them now without waiting for a further rule; that is our suggestion, not something the release asks of suppliers.
• For products that touch welfare eligibility, permits or risk scoring, document who reviews an AI output, how a human can override it, and how an affected person can seek remedy. Those are the cases the release singles out.
• Administrative-law practitioners may find the standard useful as evidence of what an agency itself says good practice is. Whether a court or tribunal would give it weight is untested in the material we opened.
• Watch for the training programme and the feedback system the release says MOIS plans, since new ethical issues would be added to the checklist over time.

What we did not verify

What we opened: the MOIS press release as republished on the Korea Policy Briefing site (korea.kr), including its embedded text, and the attached PDF of that release with the annex on the six values, on 25 October 2026.

What we did not open: the full standard and its 68 check items, any MOIS page or gazette notice, the November 2025 draft, the Ministry of Science and ICT principles the release mentions, and any statute that might give the standard legal force. The English title is our rendering of the Korean title. We did not check whether any agency has adopted the standard.

What we refuse to claim: that the standard is binding, that any penalty or deadline exists, that vendors are required to comply, or what the 68 items say beyond the six one-line descriptions in the annex.

Related briefings

• AI Regulation News hub
• Asia-Pacific AI regulation tracker
• Attorneys and AI regulation
• MSIT guide to AI research ethics in national R&D
• FSC guidance on human accountability for financial AI
• South Korea AI Act duties for high-impact AI
• Japan Digital Agency standard for government AI