Naver AI mislabels Dokdo as Japanese territory
Naver AI mislabels Dokdo as Japanese territory
Occurred: October 2025
Page published: November 2025
Naver's AI search service mistakenly identified Dokdo as Japanese territory, causing public outrage and diplomatic sensitivity in South Korea.
South Korea's largest web portal Naver's artificial intelligence search service labeled Dokdo (known as Takeshima in Japan) as part of Japan's territory in response to search queries such as “日本領土” ("Japanese territory.")
The AI also included Dokdo among Japan's major territorial claims alongside the Northern Territories and Senkaku Islands, describing the status as "disputed with South Korea."
It transpired that the content was generated based on materials from the Japanese Ministry of Foreign Affairs.
The error was flagged by Professor Seo Kyung-duk of Sungshin Women's University, who highlighted that while AIs like ChatGPT often mark Dokdo as disputed, it was unacceptable for Korea's representative portal to do the same.
Naver responded by removing the AI-generated briefings for the query and pledged to review its AI content generation to prevent future mistakes.
The incident raised concerns about the risks of AI misinformation on sensitive territorial issues and its potential impact on domestic and international perceptions of Korea's sovereignty over Dokdo.
The AI system apparently auto-ingested and summarised external public sources, in this case, Japanese government materials, without sufficient contextual validation.
There seems to have been inadequate oversight or fact-checking for politically and historically sensitive issues like territorial disputes: Naver’s AI briefly gave an unfiltered summary of the Japanese position.
There may be a lack of domain expertise in Naver's moderation or review process, meaning that responses on geopolitically disputed topics were not sufficiently curated or vetted.
For Koreans, the incident is a wake-up call that even domestic AI systems can produce historically or nationally sensitive errors. There’s a need for stronger platform review systems, especially around “hot-button” issues.
For Naver, it highlights the importance of so-called "human-in-the-loop" contextual safeguards, especially for controversial or sensitive geopolitical content, rather than relying on automatic summarisations.
For society, it shows the risk of AI amplifying unverified government positions, which could contribute to the global spread of contested narratives unless mitigated by transparency, accountability, and domain expertise.
Developer: Naver
Country: South Korea
Sector: Politics
Purpose: Summarise search results
Technology: Generative AI
Issue: Accuracy/reliability; Mis/disinformation
AIAAIC Repository ID: AIAAIC2118