As speculation spread that the "independent AI foundation model" (Dokpamo) project could be halted, South Korea's Ministry of Science and ICT moved to calm concerns. After controversy over evaluation methods, calls for frontier AI development have newly emerged, and ministry leaders are repeatedly stressing their willingness to keep the programme going.

The controversy has not fully died down. Criticism continues that the boundary between Dokpamo, which initially aimed for a world-leading level, and frontier AI is unclear.

MSIT draws a line on Dokpamo halt speculation.

Deputy Prime Minister and Science Minister Bae Kyung-hoon (배경훈) wrote on Facebook on Sept. 30, "To start with the conclusion, Dokpamo will continue," adding, "At the same time, we must challenge frontier AI." He also stressed that the third Dokpamo evaluation would be completed as planned. His comments followed a spate of reports that the government was reviewing a plan to halt Dokpamo and pursue frontier model development through steps such as setting up a special-purpose company (SPC).

Ryu Je-myeong (류제명), the ministry's second vice minister, also sought in public to address uncertainty surrounding Dokpamo. After delivering congratulatory remarks at a conference held by a media outlet on Sept. 29, Ryu mentioned Dokpamo and said, "In a rapidly changing environment, there can be various controversies and opinions about whether the path we are taking is the right direction. We will communicate with more experts and industry so that we do not miss an important moment."

The ministry said in explanatory materials on Sept. 28 that it acknowledges the need to concentrate funding to develop top-tier AI models, but said "nothing has been decided" on whether to halt Dokpamo or set up an SPC.

The controversy also spread into politics. Independent lawmaker Han Dong-hoon (한동훈) mentioned the issue on Facebook on Sept. 30 and criticised the government's AI policy. Han argued, "The environment has not changed. The strategy was lacking."

Dokpamo is a project to secure national AI competitiveness through a foundation model developed in-house by domestic companies. Its goal is to reduce reliance on global AI models and build AI competitiveness comparable to the United States and China. It is also a project the Lee Jae-myung government has pursued as a national policy task. LG AI Research, SK Telecom and Upstage are currently preparing for the next evaluation as three teams.

The initial target was also high. The ministry required elite teams to aim to challenge at least about 95 percent of the performance of the latest global models disclosed within the previous six months as of the evaluation point. It also presented a "moving target" approach that adjusts goals and methodology at each stage of evaluation. At the time, the ministry said it would "respond flexibly to the rapidly changing AI technology environment with a moving target approach."

Confusion grew after frontier AI was added as a separate axis. In the industry, there are calls for a more detailed explanation. Since Dokpamo started with the goal of developing a model close to the global top level, it is not clear how to divide goals and the boundaries of government support if frontier AI is added as a separate programme.

An industry official said, "Wasn't Dokpamo originally aimed at making a world-class top-performance model?" and added, "I wonder how it is differentiated from the frontier AI programme."

An AI industry official said, "If the nature of the project is not clear, companies will be confused about what goal they should build models for," adding, "I understand that some places are already preparing for the new frontier AI programme."

"Dokpamo is for the field, frontier is for breaking limits"... government stresses two tracks.

The ministry is stressing that the roles of the two programmes are different. Bae wrote on Facebook, "Frontier AI will break through South Korea's technological limits, and Dokpamo will become one of two pillars to make South Korea's AI used throughout industry and society."

Under the government's explanation, Dokpamo's centre of gravity would shift from a simple performance contest with global models to industrial use. Frontier AI, meanwhile, would aim for the world's top-tier models themselves by concentrating massive computing resources and domestic research capabilities.

Some analysts say the government may have played the frontier AI card because domestic models have shown a gap with the global top tier. The ministry has highlighted that South Korea is the third AI power after the United States and China, citing measures such as Stanford University's AI Index. In the "People's Sovereignty Government Year 1 MSIT Key Achievements" 발표 released in June, it also presented third place in AI model competitiveness under the Stanford index and AAII standards as a key achievement.

However, in a recent AAII evaluation, models from companies participating in Dokpamo are showing a gap with the global top models. As of Oct. 1, Upstage's "Solar Pro4" scored 28 points, SK Telecom's "A.Dot X-K2" scored 23 points, and LG AI Research's "K-EXAONE 2.0" scored 20 points. By comparison, U.S. Anthropic's "Claude Opus 5.5" scored 58 points, and OpenAI's "GPT-6 Astra" scored 53 points. China's Alibaba "Qwen 3.8 Max" also led domestic models with 45 points.

In particular, it is necessary to distinguish between AAII's assessment that South Korea ranks third in AI model competitiveness and the idea that individual domestic models have world No. 3-level performance. Domestic AI companies are forming an ecosystem by improving their capabilities to develop their own models, but the United States and China are leading in the performance competition for a single top-tier model.

Even if the frontier AI programme begins, there are many tasks to resolve. Frontier model development requires large-scale graphics processing units (GPUs), data and research manpower. If the government concentrates limited resources on frontier models, the scale and method of support for Dokpamo could change. Conversely, if large resources are投入 simultaneously into Dokpamo and frontier AI, it cannot avoid controversy over overlapping programmes.

The SPC approach is similar. It has the advantage of bringing together the capabilities of multiple companies and researchers in one place, but how to set participation criteria, the equity structure, technology ownership and the development主体 has not yet been decided. Concerns have also been raised that if the structure is built around large companies with capital and infrastructure, the space for startup participation could narrow.

Ultimately, the detailed plan the government will present in the future is expected to be the key to calming the controversy. An AI industry official said, "A declaration that Dokpamo will continue alone does not resolve uncertainty," adding, "The government needs to show a concrete blueprint for how the goals and support methods of the two programmes differ."