[Editorial] The 339-trillion-won sovereign wealth fund didn’t even check the nationality of its AI

The Korea Investment Corporation (KIC), a sovereign wealth fund that manages 339 trillion won, has adopted Chinese-made products as core models for its in-house artificial intelligence (AI), it has emerged. The models were used not only to summarize and translate documents but also to search internal materials and prepare responses to the National Assembly. They included a model developed by an institution on the U.S. government’s export-control list.
According to materials Rep. Kang Min-guk of the People Power Party received from KIC, the Chinese-made models installed in its in-house AI are Alibaba Group’s Qwen3 language model and the BGE-M3 search-embedding model from the Beijing Academy of Artificial Intelligence (BAAI). BAAI was added to the U.S. export-control list in March last year for supporting the modernization of China’s military. KIC installed the model in its in-house AI in December of the same year. The developer’s nationality and whether it was subject to sanctions were not included in the model-selection criteria.
It might seem safe, given that the system operates on a closed internal network. But it is difficult to fully understand how AI models work or what risks they pose. A closed network does not eliminate vulnerabilities in the models themselves or supply-chain security risks. An institution dealing with foreign-exchange reserves and investment strategies should assess such risks from the adoption stage. Especially as U.S.-China competition for technological supremacy intensifies, failing to adequately review the risks could also damage the sovereign wealth fund’s international credibility.
China’s National Intelligence Law stipulates an obligation to cooperate with national intelligence activities. The possibility of requests for information under relevant laws, as well as risks associated with model updates and maintenance, cannot be taken lightly. In February last year, the government successively blocked access to China’s DeepSeek on government agencies’ work networks over concerns about data leaks. It is also pursuing data sovereignty and the establishment of “sovereign AI” (AI based on domestic technology and data). If, despite this, a major financial institution managing national assets adopted Chinese models without adequately reviewing the associated risks, that would also raise questions about policy consistency.
KIC explained that it made the choice with cost and efficiency in mind, taking into account rising semiconductor prices and delivery lead times. Yet use of its in-house AI, promoted as a cost-saving measure, actually declined. Usage fell sharply from the early days after its introduction, and as of last month, ChatGPT was used 46 times as often as the in-house AI. Rep. Kang pointed out, “Performance and cost-effectiveness were considered, but the developer’s nationality and whether it was subject to sanctions were not part of the selection criteria.” Usage figures alone cannot determine whether security management is adequate, but it is worth examining whether the system has been used in a way that matches its intended purpose.
The adoption of AI by public institutions is an unavoidable trend. But models cannot be selected on price and performance alone. The government should review AI adoption across public and financial institutions and establish public AI security standards that make developers’ sanctions histories, data-processing practices, and supply-chain security essential evaluation criteria. Prior vetting and subsequent inspections should be mandatory for institutions handling sensitive information. The choice of AI by institutions managing hundreds of trillions of won in national assets cannot be judged on cost alone.