Four Services Enabled by Lower Permission-Checking Costs for Public Data

Permission Is Opening Up Before the Data Itself

The Korean government plans to open all 100 categories of public data with high demand for AI development by 2027. The schedule calls for 10 categories in 2025, 25 in 2026, and the remaining 65 together in 2027. The candidates were selected from more than 3,280 requests gathered through visits to about 800 AI companies and a public-demand survey.

Materials mentioned as possible releases range from legal interpretations and administrative appeal decisions to court cases linked to statutory provisions, Korea Fair Trade Commission resolutions, crop pest and disease photographs, and areas suitable for renewable-energy generation. They include not only documents, but also photographs, locations, measurements, and links between documents. However, the names and release dates for the 65 categories planned for 2027 have not yet been announced for each individual dataset.

The terms for using public works are also changing. After reviewing 102 categories, the government decided to move 37 categories it found free of copyright and personal-data issues into types such as the Korea Open Government License AI type (공공누리 인공지능 유형), which allows all AI companies to use them for training. Examples include institutional policy materials and publications, committee resolutions, decisions and casebooks, and some past qualification-exam questions.

Materials carrying the Korea Open Government License AI type can be modified and processed for training and used commercially. A paid service may also operate using a model trained on the material. The burden of attaching a source to every item during training is reduced, but a source must be shown when an answer directly quotes the original text. Measures must prevent outputs that are nearly identical to the original, and the compiled training material itself cannot be resold.

Previously, the cost of checking whether each document could be used for training, and asking the relevant institution when unclear, was often higher than the cost of the material itself. This made it hard for small teams to begin. The government had earlier provided 50 categories of material from 33 ministries and institutions, about 23.93 million records, to selected development teams. It now plans to extend access to the 37 categories to all AI companies. Instead of reviewing tens of millions of records one by one, teams may be able to judge the permitted scope through the type attached to each material.

What Changes at an Online Qualification-Exam Academy With 12 Staff

Consider the operator of an online academy for industrial-safety qualification exams with 12 employees. Today, staff visit multiple institutional websites, download past questions and answer files, rename the files, and put them into folders by subject. When the terms of use are unclear, they call the publishing institution or exclude that exam session from the service.

Even after collecting the questions, much work remains. Staff re-enter questions into the academy’s editor, instructors write explanations, and another employee checks whether laws and policies have changed. When a new exam session appears, they repeat the work and find and revise outdated explanations.

Sources and usage terms are the details most often missed. Individual datasets on the Public Data Portal may have different terms of use, and an AI type may be displayed alongside general terms of use. The fact that a file is available for download does not by itself mean it can be used in a paid service.

If an AI type is actually attached to past exam questions, the first task changes. Staff can gather the questions and answers with permitted-use labels and organize them in a searchable form. They can then build a feature that finds related past questions and official grounds when an exam taker enters a question they got wrong. Instead of an instructor writing every explanation from scratch, the workflow can shift to reviewing a machine-generated draft.

If policy materials and committee decisions also become available, the academy can expand its scope. Rather than stopping at the exam answer, it can connect the policy behind the question with real decision cases. Learners can check “why this is the answer” alongside official documents.

The cost reduced here is less the cost of AI development itself than the staff time spent finding materials, confirming permission, and re-entering text. One organized collection can support new-question recommendations, weak-subject diagnosis, instructor explanation drafts, and notices about revisions. The same material no longer needs to be collected again for every feature.

Work that still requires people is also clear. Instructors must check errors in questions and answers, whether revised laws have been reflected, and whether explanations are appropriate. Materials containing images or text created by third parties require separate rights checks. Do not rely only on an announcement that a material is among the 37 categories. Check the terms displayed on the actual material page and whether the conversion has been completed.

Elsewhere, Companies Charge for Ease of Use Rather Than the Data

Transport for London, London’s public transport authority, makes service information available for free and lets private companies build route-planning and transport apps. A Deloitte evaluation published in 2017 counted about 600 apps using this data and more than 13,000 registered developers. It estimated annual benefits to London residents and the transport authority at £90 million to £130 million, while the annual operating cost of making the data available was about £1 million.

In this case, private companies do not sell the underlying transport information. They earn money through alerts, route recommendations, advertising, and operational features for businesses. Even when public data is free, accurate updates and an easy interface can create separate value.

FinData in Finland accepts applications for difficult-to-release materials such as health and welfare data, then allows analysis in a secure environment. In 2025, the fee for a permit decision for a standard application from the European region was €1,600, while data combination, pseudonymization, and delivery work cost €151 per hour. It received 370 applications, made positive decisions on 91%, and processed 85% of applications within three months.

Here, users pay for permission, data preparation, and a secure analysis environment rather than downloading raw data outside. Not all public data can be fully opened, so an application-and-processing model like this may also be needed in Korea for materials containing personal data.

There is also room for services that collect parliamentary transcripts, bills, committee records, official gazettes, and regulatory-agency announcements, then provide alerts and summaries.

Customers are not paying for official documents they can read for free. They are paying for the time saved by not having to check many institutions every day. The model adds search, categorization, and alerts on top of policy materials and charges for that convenience.

Four Things You Could Build Now

1. Public Data Terms Checker

• What it does: A user enters a material URL, and the service summarizes on one screen whether AI training, paid services, modification, direct quotation, and resale are allowed.
• Who uses it: Education, legal, and mapping-service companies with five or fewer employees that want to build paid features using public data.
• Why now: Existing Korea Open Government License terms can now be displayed alongside AI types, so general use and training use need to be checked separately.
• The first screen should contain a material-URL input and a decision table written as actions, such as “training allowed,” “paid service allowed,” and “resale of materials prohibited.”

2. Evidence-Based Past-Question Study Tool

• What it does: It groups official past exam questions by subject and shows related questions, answers, and official grounds for incorrect answers.
• Who uses it: Small online academies opening new national qualification-exam courses where an AI-training permission label has been attached.
• Why now: Some past qualification-exam questions were given as examples among the 37 categories being converted, creating the possibility that collection and permission-checking costs will fall.
• The first screen should include an exam-name selector, the three most recent incorrect answers, and a button to retry related past questions.

3. Committee Decision Workflow Assistant

• What it does: It receives a question, finds relevant paragraphs and sources from resolutions, decisions, and casebooks, and summarizes them briefly.
• Who uses it: Operators at franchise headquarters that frequently receive franchisee disputes and advertising-labeling questions but do not have a dedicated legal team.
• Why now: Decisions from multiple committees are included among the examples for training-permission conversion, and Korea Fair Trade Commission resolution data is also being pursued as an opening project.
• The first screen should include a question field, three related decisions, and buttons showing the decision date and the original document.

4. Public-Institution Release Readiness Checker

• What it does: An institution uploads a document list, and the service identifies items to check for copyright holders, personal data, Korea Open Government License labels, and file formats.
• Who uses it: Communications and records staff at local public institutions that manage hundreds of publications but have only one person responsible for copyright.
• Why now: The fact that only 37 of 102 categories were selected for conversion means that institutional rights checks and label maintenance must come first.
• The first screen should include a document-list upload button and work counts divided into “ready for release review,” “rights-holder check,” and “personal-data check.”

Why this matters where you are

Korea’s planned opening schedule and Korea Open Government License AI types are Korean conditions, so their exact permissions and timing will not transfer automatically to another market. You can check whether your own public-data sources make commercial use, AI training, modification, and attribution requirements equally clear. If the rules are scattered across pages or vary by source, a service that turns those terms into a usable workflow may be worth testing.

What to Check in 30 Minutes Today

Open 10 materials in the field you want to work in on the Public Data Portal and Korea Open Government License site, then write down five items yourself: download, commercial use, AI training, modification, and attribution. If three or more of the 10 require moving between several pages to make a judgment, or if their terms appear different from one another, the Public Data Terms Checker may be worth testing as a small first service.