Decision sharpens focus on “from-scratch” standards as government clarifies independence rules
The Ministry of Science and ICT said on Jan. 15 that teams led by Naver and NCSoft have been eliminated from the government-backed competition to develop a sovereign artificial intelligence foundation model. The decision followed the completion of the first-round evaluation of the national program, which is aimed at building a homegrown AI system without relying on U.S. or Chinese technology.
Advancing to the next stage are teams led by LG AI Research, SK Telecom and Upstage. People familiar with the matter said the Naver and NCSoft groups that were cut have no plans to re-enter, leaving a smaller field as the evaluation process becomes more rigorous.
How the competition has progressed
The sovereign AI model competition began with 15 applicants and was narrowed to five finalists in August. Under the current roadmap, the remaining teams will face periodic reviews and additional eliminations, with the government planning to select two final winners by 2027.
Officials have described the program as a central pillar of South Korea’s ambition to become a top-tier global AI power, amid growing concerns over dependence on foreign foundation models and core AI technologies.
Evaluation criteria and early outcomes
The ministry said the evaluation covered a range of factors, including technical capability, policy compliance and ethical standards. Based on the first-round results, teams led by NC AI and Naver Cloud were cut from the competition, following the earlier exit of a Kakao-led consortium.
According to the ministry, LG AI Research, SK Telecom, and Upstage ranked highest overall when benchmark tests, expert evaluations, and user assessments were combined. The results drew strong interest from the industry, particularly as questions emerged over how strictly “independent development” would be interpreted.
The independence debate comes to the fore
The elimination of Naver has underscored the government’s position on what qualifies as an independent AI foundation model. Earlier this month, allegations surfaced that Naver had used vision and voice encoders, along with trained weights, from China-based Alibaba’s Qwen model. While Naver argued that its core AI engine—the “brain” of the system—was developed in-house, the government took a different view.
“In the global AI ecosystem, it is common to utilize open source,” the ministry said, adding that “the basic condition for an independent AI model is to initialize weights and conduct original training.” Using trained results developed by others, even if open source, does not meet that standard, it said.
This interpretation is likely to influence how remaining teams design their models and how future government-backed AI initiatives define technological independence.
Wild-card plan meets industry reluctance
Alongside the shortlist announcement, the ministry revealed a previously undisclosed plan to introduce a “wild-card” or repechage round, allowing one additional consortium to reenter the competition at a later stage. Teams eliminated in earlier rounds would be eligible to apply again.
Despite this option, major platform companies signaled little appetite to return. Kakao said it has no plans to rejoin the race, while Naver said it respects the ministry’s decision and will not seek reconsideration. NC AI also said it will opt out, adding that it intends to focus on industry-specific AI and other technologies that could contribute to national industries.
Concerns over momentum and incentives
Industry observers warned that the lack of interest from large players could weaken momentum behind the repechage plan. One industry official said the risk of being eliminated again, combined with unclear benefits from continued participation, may deter companies from reentering what is a resource-intensive and highly competitive process.
As the competition enters its second stage, attention is now shifting to how the government balances its push for strict technological independence with the practical realities of modern AI development. The outcome is expected to shape not only the final sovereign AI model, but also South Korea’s broader approach to defining and supporting homegrown AI in the years ahead.
