The model is a retrain of the country's existing sovereign AI, and the science ministry is separately drafting a bill on consent-free vulnerability testing.
South Korea will put a sovereign AI model specialised in cybersecurity into service by the end of 2026, Science Minister Bae Kyung-hoon said. The system is to be built by feeding security-related data into the sovereign AI model the country already operates, rather than starting from scratch.
Bae made the remarks last week at a policy briefing held at Cheong Wa Dae and presided over by President Lee Jae Myung.
The plan arrives amid growing concern over digital security, after the United States abruptly restricted exports of its advanced AI models — among them Anthropic's Mythos 5, a model built to detect vulnerabilities.
Bae said South Korea's sovereign AI as it stands lacks the capability needed to counter cybersecurity threats that keep changing and are boosted by generative AI. He also repeated earlier calls for the country to examine, over the long run, whether it should develop a frontier model on par with Mythos.
Officials at the briefing also discussed giving white hacking a formal legal footing — entering computer systems lawfully, with the owner's permission, to find weaknesses.
Separately, the science ministry is drafting a bill that would make it lawful to conduct ethical hacking against companies that have not agreed to it, for the purpose of spotting vulnerabilities.
That licence would run only in specified circumstances, and the bill is still being drafted.