With disclosure now mandatory for AI training data built under government-backed projects, a key issue has emerged over whether data companies secured at their own expense will also be subject to disclosure.

Another question is how to draw the line on what must be disclosed in projects combining government support and private investment. It is also unclear what standards the government will use if companies seek to exclude data provision on the grounds of trade secrets.

The Ministry of Science and ICT and other agencies said on Tuesday that a revised bill to the AI Basic Act, which mandates linking and providing AI training data built and produced during government projects to the national AI data platform AI Hub, passed the National Assembly plenary session on Oct. 1 and was approved by a Cabinet meeting on Oct. 13. With the law set to take effect in April next year, the government is pushing to draft subordinate regulations.

The revision aims to ensure data built with government budgets is not kept within individual agencies or companies, but can also be used by other companies and research institutions. However, data-gathering methods differ between government-commissioned data-building projects and AI development projects involving private companies, making it necessary to clearly define what falls under the disclosure requirement.

HOW FAR MUST JOINTLY FUNDED PUBLIC-PRIVATE DATA BE DISCLOSED?

A representative case is the independent AI foundation model, or dokpamo, project. In August, the Ministry of Science and ICT and the National Information Society Agency, or NIA, disclosed on AI Hub 29 types of training data, totaling about 35.44 million items, secured by five selected teams that took part in the first-stage evaluation using a 15 billion won budget for data building and processing.

Project terms at the time required at least 50 percent of the data secured using that budget to be disclosed. Naver Cloud, Upstage, SK Telecom and NC AI opened all data that had completed quality and rights verification, while LG AI Research selected data for disclosure to meet the mandatory amount.

Previously, disclosure scope was set by individual project notices and agreements, but a legal obligation to provide data will apply going forward. In projects such as dokpamo, where government subsidies and companies' own funds are both used, a key issue is which data should be considered subject to disclosure and how to reconcile project-specific disclosure terms with obligations under the law.

In particular, clear standards are needed on whether data also becomes subject to disclosure when a company uses data it already held for a government project or further processes it at its own expense.

An official at an AI data processing company said, "It may not be the case that even data commissioned at a company's own expense must be disclosed unconditionally, but the story could be different when a company further processes existing data while participating in a government project or when it invests both government subsidies and its own funds." The official added, "We are reviewing the legal principles on how far the disclosure requirement should extend."

TRADE SECRET EXEMPTIONS HINGE ON GOVERNMENT CONSULTATION AND VERIFICATION STANDARDS

Even if data falls under the disclosure requirement, it does not mean all data must be provided. The revision allows project operators and others, if they determine the case falls under reasons set by presidential decree such as protection of trade secrets, infringement of personal information or copyrights, or national security, to consult with the science minister and exclude the data from provision.

As a result, a key issue is what standards the government will use to judge requests to exclude data provision on grounds such as trade secrets. If exceptions are recognized too broadly, the obligation to open data could remain only formal. If disclosure scope is interpreted broadly, it could risk infringing companies' core technologies or data-use contracts with third parties.

The Korea Association for Trade Secret Protection also issued an official position on Oct. 13 regarding the AI Basic Act revision. While the association positively assessed that trade secrets were specified as an exception to the data disclosure obligation, it said, "Companies' establishment of pre-management systems and the government's detailing of procedures and judgment standards for disclosure exclusions must proceed in parallel."

In particular, in manufacturing AI and physical AI fields, training data may include production processes, design information and quality control know-how. This means rights relationships for data obtained jointly with partner firms, as well as companies' own information, must be considered.

The ministry plans to draft subordinate regulations in consideration of these issues. A ministry official said, "There is a six-month preparation period until the law takes effect, and procedures such as a legislative notice and vice ministerial meeting must be completed, so we believe a draft of the revised enforcement decree should be prepared within this year." The official added, "We plan to review issues raised, including the disclosure scope of jointly invested government-private data and judgments on trade secret exceptions."