A 2025 survey paper by Jonghong Jeon of the Standard Research Division, Electronics and Telecommunications Research Institute (ETRI), Republic of Korea, posted to arXiv. The paper maps international AI safety and trustworthiness standardization efforts from the perspective of South Korean government research, surveying standards produced by ISO/IEC JTC 1/SC 42, IEEE, NIST, national standards bodies, and industry consortia, and identifying gaps and proposed future directions.
Summary of argument
The paper surveys the current landscape of AI safety and trustworthiness standards and proposes future directions, framed around enhancing international competitiveness through effective standards. It treats standardization as a tool of technological competition rather than solely technical coordination, a reading the author connects to ETRI's institutional position as Korean government research and to the broader AI Sovereignty debate. The paper notes that Korea is both a major AI hardware supplier (Samsung, SK Hynix) and a significant AI user, which it presents as giving the country stakes in both the hardware and software dimensions of standards-setting.
The paper organizes the standardization challenge as escalating with AI capability. It uses a four-level capability progression: large language models (LLMs) for natural-language tasks; multimodal LLMs handling text, image, and audio; general-purpose AI (AGI), described as able to understand and solve a variety of problems like humans; and superintelligence (ASI), described as surpassing humans in all intellectual tasks. Against this progression the paper argues that current standards address LLM-era risks while standards for AGI- and ASI-level systems remain largely undeveloped.
Key concerns driving standardization
The paper identifies four categories of AI safety and trustworthiness concerns driving international standards efforts:
- Uncontrollability — AI systems that cannot be effectively stopped or corrected.
- Ethical conflicts in decision-making — algorithmic bias, fairness, and value alignment.
- Long-term socioeconomic impacts — labor displacement and concentration of power.
- Safety assurance — certification that AI systems meet safety requirements.
Standardization bodies and documents surveyed
The paper treats ISO/IEC JTC 1/SC 42 as the primary international AI standards body and reviews its key standards: ISO/IEC 42001 (AI management systems), which sets requirements for organizations using AI; ISO/IEC 23894 (AI risk management guidance); and ISO/IEC 24028 (overview of trustworthiness in AI).
For IEEE, the paper covers the IEEE P7000 series of ethics-related standards for autonomous systems and IEEE 2089 (standards for ethical AI and autonomous systems).
For NIST, the paper covers the AI Risk Management Framework 1.0 (2023), built on the Govern, Map, Measure, and Manage functions, and NIST AI 600-1, which addresses generative AI risks.
Gaps identified
The paper identifies several gaps in the current standards landscape:
- No internationally agreed standards exist for AGI-level systems; standards are catching up to LLMs rather than future capabilities.
- Fragmentation across bodies — ISO, IEEE, NIST, EU, and national standards produce different frameworks that companies must reconcile.
- Technical standards lag behind regulatory requirements: the EU AI Act references standards bodies for compliance, but many referenced standards are still in development.
- A gap exists between technical trustworthiness standards and governance and accountability standards.
Relationships
- related: NIST AI RMF 1.0 — the survey places NIST as one node in a larger international standards ecosystem.
- related: Frontier Compliance Framework — Anthropic's FCF references ISO 42001 and NIST 800-53 among its foundational standards.
- related: EU AI Act — the Act defers to standards bodies for technical implementation, and this paper maps what those bodies are producing.
- related: AI Safety Frameworks — the paper positions international standards (ISO, IEEE) alongside voluntary frameworks (RSP) and compliance frameworks (FCF).
- related: AI Sovereignty — the paper frames standards competition as a form of technological competition and presents Korea's participation as an example of how non-superpower nations exert influence in AI governance.
- related: OECD AI Report — the OECD governance framework is operationalized through the standards-body processes this paper surveys.
Provenance
Jonghong Jeon, Standard Research Division, Electronics and Telecommunications Research Institute (ETRI), Republic of Korea. Published on arXiv, 2025. (Source: Raw Sources/Standardization Trends on Safety and Trustworthiness Tech for Advanced AI.md)