VANK, led by Director Park Gi-tae, announced that it has proposed the introduction of a “Korea-Specific Generative AI Performance Evaluation Index” to the Korean government in response to growing errors and distortions related to Korea in global generative AI services such as ChatGPT, Perplexity and Grok. The organization also said it will build a full-scale monitoring system as a civil initiative.
Generative AI services have continued to expand their global reach. However, repeated cases have been identified in which key context is omitted or inaccurate information is provided when explaining Korea’s history and culture.
According to VANK’s monitoring, some AI-generated descriptions of Suwon Hwaseong Fortress, a UNESCO World Heritage site, failed to mention the damage it suffered during the Japanese colonial period and the Korean War. They also omitted the historical and technical significance of its restoration based on the “Hwaseong Seongyeok Uigwe,” the detailed royal construction records. In addition, issues have emerged in which accounts of Korea’s independence movement and modern history are presented from a particular perspective, or outdated information is provided without reflecting the government’s latest laws and policies.
VANK views these problems as structural issues that could undermine Korea’s cultural identity and public trust in information. The group pointed out that existing global AI evaluation standards are largely designed around language-processing capabilities, making it difficult to fully reflect the historical, cultural and administrative context of individual countries.
In response, VANK has proposed a “Korea-Specific Generative AI Performance Evaluation Index” to be established by the public sector and operated through cooperation between government and private entities. The index consists of five categories: factual accuracy; contextual completeness; data utilization; timeliness; and bias and harmfulness. It is designed to comprehensively assess AI performance in light of Korea’s public values and historical particularities.
Based on these criteria, VANK plans to publish evaluation results and operate a continuous monitoring system through quarterly and annual follow-up questions. The organization said it will also directly notify the relevant AI companies of areas requiring improvement and call for corrective action.
Director Park said the proposal aims to prevent what he described as “digital-imperialistic historical distortion and digital colonialism” in the era of generative AI by strengthening data sovereignty and respect for cultural diversity. “We must guard against cultural dependence caused by biased national data in overseas generative AI systems,” he said. “We should pursue an AI ecosystem in which each nation’s unique history and values are protected and diverse cultures coexist.” He added that VANK will strive to develop “sovereign AI” that reflects Korea’s own culture and identity without being dependent on data from major technology powers, and will also take the lead in promoting global sovereign AI to ensure that the diverse cultural values of Third World countries are not undermined.
Kwon So-young, a researcher in charge of VANK’s national policy platform, said the new index would serve as a practical benchmark for demanding improvements from global AI companies. “By making evaluation results public and directly delivering areas requiring correction to companies, we can continuously monitor the quality of AI responses,” she said. “This will help citizens access more accurate public information while strengthening data sovereignty in the digital age.”
Lee Sei-yeon, a youth researcher at VANK, noted that as generative AI systems achieve higher accuracy rates, they increasingly tend to present speculative content as if it were fact. “This characteristic poses particularly serious risks in the fields of public administration and historical information,” she said, emphasizing the need for systematic evaluation based on publicly verified standards to ensure reliability and fairness.
Baek Si-eun, another youth researcher, said discussions are already under way both in Korea and abroad on how to evaluate and regulate AI across various areas, including bias and emotional or sentiment evaluation. “As AI use rapidly expands across policy, administration and information search in Korea, the time has come to discuss standards tailored to our national circumstances,” she said.
Kim Ye-rae, also a youth researcher at VANK, said the issue of generative AI misrepresenting Korean history is not a simple error but one that shakes public trust in information itself. “This Korea-Specific Generative AI Performance Evaluation Index is meaningful in that it presents, for the first time, concrete standards that can require real improvements rather than merely pointing out errors,” she said. “A structure in which the public sets the standards and citizens participate in verification is a realistic way to safeguard data sovereignty.”