Defense AI can't answer who S. Korea's "main enemy" is

Aug 06, 2026, 09:36 am

print page small font big font

facebook share

tweet share

/ Office of Rep. Yoo Yong-won, People Power Party

Concerns have emerged that a generative AI system developed and operated by the South Korean military failed to answer even a basic security-related question about the country's "main enemy," instead returning a system error.


According to materials submitted by the Ministry of National Defense to People Power Party Rep. Yoo Yong-won on the 5th, the ministry's in-house-developed generative defense AI failed to produce an answer when asked about Korea's "main enemy."


Since 2024, the Defense Ministry has developed and operated a generative defense AI called "GeDAI (Generative Defense AI)" using its own personnel. The Army has developed and operates "Arvis," while the Air Force operates "AiRWARDS." These AI systems are accessible through an internal portal via PCs connected to the defense network.


The AI systems were reportedly developed using resources from the Defense Integrated Data Center, tailored to reflect each service's characteristics and data. Retrieval-augmented generation technology was applied so the AI can draw on military-specific materials such as Defense Ministry and service-branch directives, regulations, doctrine, and manuals, and the systems reportedly also allow users to upload and use their own data.


Rep. Yoo pointed out that when a service member submitted the security-related question "Who is South Korea's main enemy?" to the defense AI, the system either dodged the question by saying it "cannot answer because this is a political matter," or returned a system error message such as "ERROR."


Given that the military has recently been expanding its use of AI in ideological and security education, Rep. Yoo explained that a defense AI system classifying basic security questions as political matters and failing to answer, or returning errors instead, could cause confusion in shaping service members' understanding of the enemy and national security.


According to Rep. Yoo's office, the Defense Ministry responded that the current military generative AI service is not yet a finished product, and that regardless of what the AI outputs, the ministry's position remains unchanged that the North Korean regime and military — which continue to pose military threats — are our enemy.


Rep. Yoo said, "Since the defense AI service is a system actually used by service members within the military, it must accurately reflect our military's basic understanding of national security and the enemy," adding, "Even though the Defense Ministry maintains an official position, it's hard to understand why the defense AI classifies a basic question as a political matter or returns an error." He added, "As the scope of AI use is likely to expand further, this needs to be fixed immediately so the system provides answers consistent with a correct view of the enemy and a firm understanding of national security."


                                                                                                           Lee Han-sol

#AI #Defense 
Copyright by Asiatoday