The government will speed up efforts to build an agentic AI ecosystem in which AI plans on its own and carries out tasks. It is based on an assessment that competition in AI technology is shifting to execution capabilities that connect real services and systems.

The Ministry of Science and ICT on Wednesday announced a joint interagency "Agentic AI Initiative" at the 11th Ministerial Meeting on Science and Technology Affairs at the Government Complex Seoul. The meeting was chaired by Vice Prime Minister Bae Kyung-hoon (배경훈), who also serves as science minister.

Agentic AI is a technology in which AI autonomously draws up plans and uses external systems and tools to execute tasks. The government said the axis of AI competition is expanding beyond securing strong models to ecosystem competition that connects and executes data, systems and work processes.

Under the initiative, the government will pursue three strategies: leading safety and trust for agentic AI, building an open foundation for agentic AI execution, and promoting demand-led adoption of agentic AI.

RIGHTS ABUSE, DATA LEAKS IN FOCUS AS AGENT MARKETPLACE PLANNED

The science ministry will establish a safety and trust management framework to respond to new risks such as abuse of authority by AI agents, security threats and leaks of sensitive information. It will prepare agentic AI safety and trust guidelines within this year covering protection of personal and sensitive data, AI authority management and prevention of goal deviation.

It will also build datasets, scenarios and benchmark tools to evaluate the performance, safety and trustworthiness of agentic AI. To ensure accountability for AI agent actions, it plans to review measures for agent identity verification and management, as well as where responsibility lies in case of accidents.

The government will build an open execution ecosystem suited to domestic conditions. It will support the development of AI agent services and tools closely tied to everyday life and in high demand, and push to create a private marketplace that connects and distributes verified agents and services.

It will also develop core technologies for an "AI operating system" that integrates management of AI models and inference systems, agents and external tools. It will work to secure multi-agent orchestration technology that coordinates the tasks of multiple agents, as well as long-context planning and self-correction. From 2027, it plans to support graphics processing units and data to strengthen the agentic AI performance of locally developed AI models.

"AI FOR ALL" SEEKS ONE-AGENT-PER-PERSON ERA WITH INDUSTRY FLAGSHIP PILOTS

The government will also expand the adoption of AI agents. Through the "AI for All Project", it will lay the groundwork for anyone to use a personal AI agent. In the long term, it will support infrastructure costs in the form of vouchers so individuals can use AI agents to handle tasks or pursue one-person startups.

In key industries including media, logistics, fintech, edtech, legal tech, energy and healthcare, it will form large consortiums involving AI model companies, infrastructure firms, service developers and user companies to support development and demonstration of agentic AI services.

To help companies develop and operate AI agents on their own, the government will support introduction diagnostics, risk management and establishment of operating systems, and build an "AI agent industry map". For services that are difficult to test under existing regulations, it will apply the ICT regulatory sandbox to expand testing opportunities.

The science ministry will flesh out detailed policies in the second half of the year, centered on a joint public-private "Agentic AI Alliance". About 280 companies are participating in the alliance, which will discuss tasks by area including industry, technology, ecosystem, safety and trust.

MEETING ALSO DISCUSSES SMR AND DEFENCE SEMICONDUCTOR PLANS

The meeting discussed six agenda items including the agentic AI initiative, directions for fostering next-generation small modular reactors, a localisation strategy for defence semiconductors, a physical AI port strategy for Jinhae New Port, a public-sector AX privacy protection guide and GPU support plans for major government AI projects.

The government will provide 512 B200 GPUs to the "AI for All Project". It will also allocate 8 GPUs to a public AI legal statute search service. It will also push to build a next-generation SMR supply chain, create a physical AI port at Jinhae New Port, and establish an inspection system to prevent privacy violations during AI transformation at public institutions.

Bae said the second year of the "people's sovereignty government" is "a time to prove our direction through achievements" amid a golden time for a major leap. He stressed that all related ministries and agencies should join in active implementation so the public can feel the change.