Global AI policy watch

Overseas Trends

Identifies cases where AI is used for policy problem definition, evidence analysis, opinion gathering, alternative development, and evaluation in official materials from governments and international organizations, and summarizes them in Korean.

RSS Subscription Suggest Trends

Current publications and automation readiness are reported separately.

The eight currently published items were researched and edited by Codex from official sources; the exact underlying model version was not recorded at the time. The Upstage solar-open2 private beta, made available with priority access support after participation in an Upstage event, completed authenticated calls and produced one unpublished duplicate draft. The Global Trends source-comparison step using nvidia/nemotron-3-ultra-550b-a55b:free and scheduled automatic publication have not yet been activated.

Japan · 2026-07-17

Japanese Government AI 'Gennai' Supports Verification of Policy Hypotheses from 8,000 Agricultural Policy Survey Responses

Japan's Ministry of Agriculture, Forestry and Fisheries established over 20 policy hypotheses based on rice production intention surveys, and the Digital Agency reviewed the validity of the hypotheses from approximately 8,000 responses using AI. Official explanations state that analysis taking one person about two months was reduced to about 3 days.

Utilization in Policy Development: Used in the stage of repeatedly verifying hypotheses in large-scale surveys after defining policy problems, and in evidence and opinion analysis to classify public comments.
Human Review: Public officials are responsible for setting policy hypotheses and verification instructions, while AI supports analysis. It is not described as a case of automating final policy judgment.
AIPOL Implications: A direct reference case applying AIPOL's 'Policy Hypothesis → Large-scale Response Verification → Human Judgment' flow to actual administrative data.
Limitations: The official page lacks sufficient disclosure of details regarding individual models, accuracy, bias checks, sample representativeness, and error correction procedures.

Codex AI Investigation and Editing Summary · Japan Digital Agency Original Text ↗

International · 2026-06-01

OECD: AI Adoption Spreading in Government, Yet Policy Development Use Remains Limited

The OECD Digital Government Outlook summarizes that while most surveyed countries use AI in some government functions, adoption in high-risk areas such as policy development and oversight remains relatively limited. Data quality, transparency, assurance, organizational capacity, and continuous stakeholder engagement are identified as key conditions for expansion.

Policy Development Use: Comparison of AI Adoption Status Across Government Policy Design, Decision-making, and Services
Human Review: Integrating Institutional Governance and Internal/External Stakeholder Engagement Throughout the AI Adoption Process
AIPOL Implications: The AIPOL case provides grounds for designing initiatives not merely as technology demonstrations, but alongside verification, accountability, and participation structures.
Limitations: Institutions and survey scopes vary by country; therefore, numerical values between countries should not be interpreted as simple rankings.

Codex AI Research and Editing Summary · OECD Original Text ↗

Poland·EU · 2026-03-24

Poland Unveils Government AI Assistant Pilot to Aid Regulatory Impact Assessment Writing

The Regulatory Impact Assessment Department of the Polish Prime Minister's Office unveiled a pilot intergovernmental virtual assistant using the Polish language model PLLuM and Retrieval-Augmented Generation (RAG). It supports policy officials in finding regulatory information and guidelines and preparing impact assessment documents.

Policy Development Application: Applied as an interactive assistant tool in the stages of exploring regulatory grounds, checking guidelines, and preparing impact assessments and document writing.
Human Review: Introduced as a support tool for regulators; does not replace impact assessments or policy decisions.
AIPOL Implications: Demonstrates a reusable structure for converting a regulatory impact assessment knowledge base into a RAG-based policy development workspace.
Limitations: Currently a pilot; performance metrics, error rates, and impact on actual regulatory decisions have not been disclosed on the official event page.

Codex AI Research & Editing Summary · European Commission Interoperable Europe · Polish Prime Minister's Office Original Text ↗

UK · 2025-12-23

Department for Transport Publishes Technical Evaluation of AI Public Opinion Analysis Tool

The UK Department for Transport and the Alan Turing Institute published an evaluation comparing CAT's performance in topic-analyzing policy consultation responses against human benchmarks. Key focus: evaluating user research, technical performance, and human oversight together to handle citizen opinions accurately and responsibly.

Policy Development Application: Topic Analysis of Free-Text Responses in Policy Consultations
Human Review: Evaluate against human benchmarks and maintain oversight by policy analysts
AIPOL Implications: This case demonstrates that policy development AI must be evaluated not only on outcomes but also on human baselines, error types, and operational procedures.
Limitations: Evaluation results for a specific tool cannot be directly generalized to other languages or policy fields.

Codex AI Research & Editing Summary · UK Department for Transport Original Text ↗

Singapore · 2025-12-01

Singapore Urban Planning: Visualizing Citizens' Spatial Ideas with AI Images

Singapore URA operated an AI-based Dream Lab during the 2025 Master Plan public engagement process. Exhibition visitors visualized ideas for 16 future development sites as generated images, and these were utilized alongside other engagement methods such as surveys, workshops, and dialogues to review long-term land use plans.

Policy Development Application: Assisted citizens in expressing abstract spatial policy ideas as visible alternatives, utilized in co-design and opinion gathering stages.
Human Review: AI handled idea visualization, while surveys and workshops operated by staff and volunteers, and institutional planning assessments were conducted in parallel.
AIPOL Implications: A participatory design case converting policy scenarios into image prototypes for citizens to compare and discuss concretely.
Limitations: It is difficult to verify the impact of individual AI images on the final plan, the models and safety measures used, and the representativeness of participants based solely on official announcements.

Codex AI Research·Editing Summary · Singapore Urban Redevelopment Authority (URA) Original Text ↗

World Bank · 2025-10-17

World Bank ImpactAI, linking causal research to policy alternative comparison materials

World Bank DIME AI's ImpactAI is a policy evidence tool that queries and synthesizes a public causal research database centered on randomized controlled trials using generative AI. It enables comparison and visualization of results by intervention and allows users to navigate to original source evidence.

Policy Development Utilization: Supports evidence search and synthesis aligned with policy goals, comparison of intervention alternatives, and resource allocation decisions.
Human Review: Policy officials must verify summaries and comparisons by linking to original sources, retaining final decision-making on alternatives and contextual judgment.
AIPOL Implications: A strong reference case for evidence-to-options design connecting traceable evidence databases to policy option development.
Limitations: Current public explanations focus on development economics and English-speaking randomized controlled trials, excluding qualitative context, institutional differences, and operating mechanisms.

Codex AI Research and Editing Summary · World Bank DIME AI Original Text ↗

UK · 2025-05-14

Consult: Government AI tool analyzing public consultation responses

The UK Government's Consult tool identifies themes and patterns from large-scale free-text responses submitted to policy consultations for analyst review. Developed internally by the government, algorithm transparency records were published alongside it from the pre-release testing and evaluation stage.

Application in Policy Development: Initial Theme Analysis and Exploration of Public Consultation Responses
Human Review: Policy staff verify and correct AI-identified themes and response classifications
AIPOL Implications: Demonstrates a structure that enables rapid review of numerous opinions while reserving judgment on policy implications for responsible officials.
Limitations: AI does not determine response representativeness or policy direction; the tool was in the pre-deployment stage at the time.

Codex AI Research & Editing Summary · UK Department for Science, Innovation and Technology Original Text ↗

UK · 2025-04-23

UK Government Compares AI-Supported Evidence Review and Human-Only Review on the Same Task

UK DSIT and DCMS reviewed the same policy question using AI-supported and human-only methods respectively. The AI-supported review reduced total time by 23% with similar final quality, though initial drafts were less fluent and required more editing.

Application in Policy Development: AI tools were used for literature search, screening, analysis, synthesis, and deriving policy implications, comparing time and quality.
Human Review: Researchers manually verified and edited AI outputs, and seniors reviewed quality. Auto-generated content was not used directly as policy evidence.
AIPOL Implications: A methodological reference case evaluating AI policy research tools not only on speed but also on quality, editing costs, and errors through controlled experiments.
Limitations: Hallucinations and link errors occurred; as this is a comparative study on a single topic, results cannot be generalized to other policy areas or tools.

Codex AI Research & Editing Summary · UK Department for Science, Innovation and Technology (DSIT) Original Text ↗