Credibility slips in national AI drive

The government’s so-called national AI selection appears to have drifted off course. The competition began with five teams, but a leading contender was soon entangled in allegations of plagiarizing Chinese technology, and two teams were eliminated instead of the one initially planned. The confusion did not stop there. Officials said they would add another team through a repechage round, yet no major company stepped forward. With domestic tech giants such as Naver, Kakao and KT opting out, only two startup-led consortia submitted proposals. Momentum is ebbing behind the “sovereign AI foundation model project,” originally conceived to cultivate globally competitive artificial intelligence.

The turmoil was hardly accidental. It exposed structural flaws embedded in the project’s design. The Ministry of Science and ICT adopted a survival-style format to capture public attention and spur corporate rivalry. Yet in a state-led effort to designate a national AI champion, elimination carries consequences that extend beyond procedure. It inflicts reputational damage. For companies, the contest became less a measure of technological capability than a test of brand risk. Few firms would willingly assume such exposure.

The requirement that a model be built “from scratch,” in other words developed entirely independently, has further intensified the controversy. Authorities have said qualifying models must be free from any licensing disputes involving other companies and have made clear that systems created by fine-tuning overseas models would not be recognized as domestic. However, they failed to clarify how extensively foreign open-source technologies could be utilized. In a government-led initiative, such ambiguity quickly becomes a source of risk.

Repeated changes to the rules further weakened public confidence. The government first indicated that only one of the five teams would be cut, only to disqualify two in the end. It later said it would bring in an additional team to retain a four-team structure. As the framework moved from one elimination to two and then to an added selection, the credibility of the “national representative” label was bound to suffer.

The government’s dilemma is understandable. With the United States and China locked in a state-level race for artificial intelligence supremacy, South Korea may have felt compelled to launch a flagship initiative of its own. The survival-style format likely aimed to generate public interest despite its inherent risks. The urgency to prove the competitiveness of domestic AI models is also evident. Yet a national champion that lacks public confidence cannot command recognition at home, let alone compete abroad.

As the second round of evaluations approaches, this is the moment to pause and recalibrate. Selecting a national AI champion is not a television audition. An audition concludes with ratings; a national AI strategy shapes long-term competitiveness. What is needed now is not promotion but an honest reckoning with the sources of repeated confusion. Trust must outweigh spectacle. Officials should clearly spell out how evaluation timelines and criteria will apply to both the existing second-round teams and any additional entrants, and establish transparent, consistent standards for verifying model independence if the issue arises again.

Choosing a national AI matters. Ensuring that the choice commands broad recognition matters even more. The priority now is not speed but trust.