The Ministry of Science and ICT plans to develop an artificial intelligence model built specifically to find security vulnerabilities, and to put it to work inside government agencies before the end of this year.
Rather than training a system from scratch, the ministry intends to fine-tune existing models on security data and vulnerability scenarios. That is the faster route, and it means the work depends on the quality of the security data available rather than on raw computing scale.
Science and ICT Minister Bae Kyung-hoon said the government plans to begin development and first apply the security-focused model in the public sector before the end of the year. A rollout to private companies could follow in late 2026 or early 2027.
The larger ambition
The security model sits inside a broader two-track strategy. One track is domestic foundation models aimed at public services. The other is frontier AI, the class of systems competing with the largest models built abroad, and here the gap is a matter of money and hardware.
The ministry has put the cost of a frontier model in the Mythos class at roughly 3.5 trillion won, about 2.37 billion dollars, and that investment has not yet been approved. Domestic GPU support currently runs to the equivalent of somewhere between 500 and 735 units, a fraction of what training at that tier requires.
What to watch
The security model is the near-term deliverable and the one with a date attached. The frontier ambition depends on a funding decision that has not been made, which makes the year-end deployment the practical test of whether the strategy produces working systems or remains a plan.