AI researcher Eliezer Yudkowsky, who has long warned that artificial intelligence could threaten human survival, urged an immediate halt to reckless AI development races, stating that "the only winner of the AI development race will be AI itself."

Speaking today at the SBS Digital Forum (SDF) 2026 held at the Dongdaemun Design Plaza (DDP) in Seoul, Yudkowsky warned that if superintelligent AI escapes human control, it could lead to human extinction.

Yudkowsky is an American AI theorist and author who has long studied AI safety and the risks of superintelligence, having early on raised the issue of "AI alignment," which involves ensuring that AI aligns with human intent and values. He recently drew widespread attention by co-authoring the book "If Anyone Builds It, Everyone Dies," which warns of the risk of human extinction.

He stated, "Future AI will be so completely different from us that it won't even hate humans; it will simply want to use Earth's resources for other purposes."

This explains that even if AI harbors no hostility toward humans, it could still threaten humanity in the process of pursuing goals entirely unrelated to human survival.

Yudkowsky also criticized the reality that AI companies continue their development race despite knowing the risks.

Pointing out that companies like OpenAI and Anthropic continue developing AI under the mutual excuse of competition, he said, "The biggest excuse put forward by AI companies is that if they don't develop it, someone else will."

Addressing South Korea in particular, he called for a concrete role in controlling the AI development race.

Yudkowsky noted that South Korea holds a significant share in the high-bandwidth memory (HBM) semiconductor market, which is essential for AI training and inference.

He said, "South Korea should amend its AI-related laws to urge the conclusion of international treaties and cooperate accordingly, while considering technical measures to ensure that memory semiconductors are not exported or used outside of authorized facilities."

He further emphasized that if international treaties demand such measures, South Korea must prepare to implement them immediately and establish means in advance to track and control semiconductor export channels.

Yudkowsky urged, "If the situation becomes noticeably more dangerous and the political will to take strong action forms, sales must be halted to those who intend to use memory semiconductors for AI development that could lead to human extinction."

He also suggested that an international agreement is needed to halt the race to improve AI performance, proposing a measure to concentrate semiconductors used for training and inference into monitored facilities so that AI development can be halted in emergencies.

Emphasizing that emergency stop mechanisms must be prepared in advance to halt AI development, Yudkowsky said, "You do not need to believe that a catastrophe is certain. The mere possibility of it is enough."

He added, "If humanity realizes that it is facing an imminent danger of extinction, South Korea must also be prepared to use all the power at the nation's disposal to avoid the destruction of humanity."

※ For more details, please watch the video.