appConfigEntityId

bodyType

customContents

entityId

entityUuid

isVideo

articlePageQuery

article

articleSchemaContent

content

hooksArticlePageAdvertisementConfigArticle

hooksArticlePageFloatingButtonConfigArticle

appConfig

filter

json

liveArticle

Park Chan-kyong

South Korea plans to develop its own cybersecurity-focused artificial intelligence model by the end of this year, after a brief US clampdown on advanced systems exposed the risks of relying on foreign technology for national cyber defence.
But Seoul’s more ambitious, longer-term goal of building a frontier model capable of competing with the world’s most advanced systems would require it to overcome a persistent gap in AI software, computing capacity and large-scale training, analysts said.
The initiative reflects a broader push for “AI sovereignty”, with governments in Japan, Britain, Canada and across the European Union investing in domestic models and computing infrastructure to reduce their reliance on foreign providers.
Science and ICT Minister Bae Kyung-hoon unveiled the plan on Thursday after Washington imposed new export restrictions on several advanced AI models, including Anthropic’s Mythos 5.
“We are pushing to create an AI model specialising in cybersecurity within this year by training our existing sovereign AI models on security-related data,” Bae said during a policy briefing chaired by President Lee Jae Myung.
Bae warned that AI-powered cyber threats were becoming increasingly sophisticated.
“AI can be used to identify security vulnerabilities and exploit them with ease, so we need to prepare our defences accordingly,” he said.


Although several South Korean companies have developed their own large language models, Bae acknowledged that none yet matched the capabilities of the world’s leading frontier AI systems.
“They are not yet capable of responding effectively at that level,” he said.
Bae outlined a two-pronged strategy: build a cybersecurity-focused AI model in the short term by further training existing sovereign AI models on security-related data, then explore developing a frontier AI model comparable to Anthropic’s Mythos.
“Although Mythos was not developed specifically for cybersecurity, it has become capable of solving complex security problems with ease,” he said. “We therefore need to seriously consider developing similarly advanced frontier AI models.”
The policy briefing also focused on uncertainty surrounding US restrictions on access to advanced AI systems.
President Lee said South Korea must prepare for the possibility that access to foreign AI models could be suspended.
“We need to assume that access could be blocked,” Lee said. “We cannot allow a situation in which another country is effectively protecting our cybersecurity and then suddenly leaves the gate.”
Bae echoed those concerns. “The United States restricted access to advanced AI, then relaxed the restrictions and has recently reintroduced limited controls. China appears to be moving in a similar direction, so we don’t know when access could be cut off,” he said.
Lagging behind
A senior Science Ministry official estimated that South Korea trailed the US by about six months in AI technology.
“The government has been supporting five selected pioneers by providing them with AI computing resources, including GPUs, and data to develop our own frontier AI model,” the official, who asked for anonymity, told This Week in Asia, noting that the government had been pursuing a sovereign AI foundation model project since August 2025.

Professor Kim Jang-hyun, an AI expert at Sungkyunkwan University, said South Korea’s weakness was in software rather than hardware.
“As far as AI software is concerned, South Korea is considerably behind not only the United States but China as well,” Kim said.
One visible sign of that gap, he said, was the lack of widely adopted domestic generative AI platforms capable of competing internationally with ChatGPT, Google’s Gemini, Anthropic’s Claude or China’s DeepSeek, although companies including Naver, LG and Upstage have developed their own models.
Kim attributed this partly to years of underinvestment.
“In the past, the Korean government was not very enthusiastic about investing money and resources in developing software that did not generate tangible short-term returns, while local companies had to compete within a relatively small domestic market,” he said.
“South Korea has the human resources and technological capability to develop a top-tier AI. It depends on how much the government is willing to invest,” Kim added. “I wouldn’t be surprised if it produces a frontier AI capable of competing with the world’s leading systems within a year or two.”
Another AI specialist working at a Korean technology company was less optimistic.
“There are still extremely difficult final stages in developing generative AI, including large-scale training, evaluating outputs and integrating models into enterprise workflows. No one can say with confidence how long it will take South Korea to overcome those challenges,” said the specialist, who asked for anonymity as he was not authorised to speak to the press.
Kim said South Korea’s push to develop security-focused frontier AI went beyond technological competition. “It represents an extension of the country’s broader effort to reduce its dependence on foreign AI giants and secure national AI sovereignty.”
While the global AI landscape has largely been dominated by the US and China, South Korea has relied heavily on overseas companies for foundation models, cloud infrastructure and AI software, exposing itself to risks should foreign governments or companies restrict access to critical technologies, according to Kim.
“Although South Korea may have entered the race later than some competitors, it must now pursue broad research and development initiatives to strengthen its domestic AI capabilities rather than relying unilaterally on foreign technology companies,” he said.
Kim said AI-powered cybersecurity should become a national strategic priority as generative AI and autonomous agents created new security risks.
AI could already automate malware development, vulnerability discovery, phishing campaigns, social engineering attacks and attack planning, he said.

“Looking ahead, fully autonomous cyberattacks – where AI independently plans and executes multiple stages of an attack – are increasingly becoming a realistic possibility,” he warned.
Against that backdrop, Kim said governments should be cautious about relying heavily on foreign AI systems to protect critical national infrastructure.
“No matter how capable a security solution may be, if the government cannot fully control how data is processed, how the model operates or how updates are managed, it effectively entrusts critical national information and security infrastructure to external entities,” he said.
Rather than excluding foreign technologies altogether, Kim said South Korea should build domestic capabilities while continuing to use overseas technologies where appropriate.
“The objective is not to reject foreign technology,” he said. “It is to ensure sovereign control over critical data, cybersecurity decision-making and essential infrastructure.”
He added that achieving that goal would require long-term cooperation among government, industry, universities, research institutes and local authorities.
“A country that lacks the ability to choose and control the AI technologies underpinning its critical systems cannot truly be considered an AI leader,” Kim said. “Developing security-centred frontier AI is therefore a strategic imperative that advances both industrial competitiveness and national security.”


Suspected landmine blast injures South Korean soldiers in DMZ

Japan asks for US$43 million to weed out cherry tree-devouring longhorn beetle

South Korea steps up to plug Europe’s jet fuel deficit

US announces AI dialogue with China, launches Board of Trade

South Korea

Why South Korea is racing to build a sovereign AI model for cybersecurity

Officials at the Korea Internet & Security Agency (KISA) check internet systems at the KISA situation room amid reports that several computers were struck by a new type of ransomware the previous night, in Seoul, South Korea, on June 28, 2017. Photo: EPA

The Anthropic website and its logo. Washington imposed new export restrictions on several advanced AI models, including Anthropic’s Claude Mythos 5. Photo: AP

South Korean President Lee Jae Myung speaks during a meeting in Jinju, South Gyeongsang province, on July 3. Photo: EPA

Southeast Asian scammers’ new disguise: your leader’s face

Illustration: Kevin Wong

AI-generated deepfakes of Southeast Asian leaders are increasingly being used by online scam syndicates. This illustration was produced with the assistance of Google’s Nano Banana. Illustration: Greg Fountain

On criminal forums and Telegram channels, chatter about deepfakes surged 600 per cent in just one five-month period in 2024. Photo: Reuters

“Cybercrime exists wherever anyone is using a smartphone or laptop”, law enforcement officials say. Global losses hit US$10.5 trillion in 2025. Photo: Shutterstock

AI and other tech tools are increasingly being integrated into scam operations across Southeast Asia and further afield, experts say. Photo: Shutterstock

An emergency warning alert alarm on a mobile phone. Artificial intelligence has not built a new criminal underworld, it has retooled an old one – at speed. Photo: Shutterstock

Why South Korea’s submarine exports are running aground

South Korean submarines take part in a naval fleet review off the southern port city of Busan last year. Photo: AFP

Canada’s Prime Minister Mark Carney announces the winning bid to make 12 submarines for the country’s navy on July 6. Hanwha Ocean was named as the “reserve supplier”. Photo: The Canadian Press/AP

Hanwha Ocean’s shipyard in Geoje, South Gyeongsang province, South Korea. Photo: Korea Times

South Korea’s Dosan Ahn Chang-ho, a KSS-III submarine that was first launched in 2018. Photo: Korea Times

Flags of member states flap in the wind outside Nato headquarters in Brussels earlier this year. Photo: AP

South Korea’s Shin Chaeho submarine, launched in 2021, takes part in a naval fleet review last year. Photo: AFP

South Korean ticket touts to face the music when new penalties come into play

K-pop group BTS performs in central Seoul on March 21. Photo: Reuters

Boy band Seventeen greets the audience in New Jersey on September 7, 2022. Photo: Pledis Entertainment

South Korean boy band NCT Dream becomes the first K-pop group to perform at Hong Kong’s Kai Tak Stadium in August 2025. Photo: Facebook/NCT Dream

Boy band Stray Kids appear on the red carpet at the 2022 MAMA Awards in Osaka. Photo: AFP

This Week in Asia

Politics

News

Asia

East Asia

Fears over access to foreign AI models have accelerated Seoul’s push to control the systems used to defend critical infrastructure


Fears over access to foreign AI models have accelerated Seoul’s push to control the systems used to defend critical infrastructure.


Why South Korea is racing to build a sovereign AI model for cybersecurity

.css-1c6uqr6{color:inherit;font-weight:inherit;font-size:inherit;font-family:inherit;line-height:inherit;overflow-wrap:break-word;}Fears over access to foreign AI models have accelerated Seoul’s push to control the systems used to defend critical infrastructure

Fears over access to foreign AI models have accelerated Seoul’s push to control the systems used to defend critical infrastructure