South Korea races to build its own security AI model

South Korea is developing its own security-focused AI model and wants it online by the end of 2026, aiming to give the country sovereign bug-finding capabilities instead of relying on systems controlled abroad.

Deputy Prime Minister and Minister of Science and ICT Bae Kyung-hoon disclosed the plan and said South Korea needs a model that can rival Anthropic’s Mythos. According to The Register, the push follows two US moves that cut off access to Mythos: first, a requirement that Anthropic offer it only to American citizens, which the company could not do and so blocked all access; and second, an order to take down the service while Washington investigated allegations of potentially dangerous performance problems.

Those episodes, the report says, convinced many governments that the US could deny access to powerful models in the future, leaving US-based organizations and national security agencies with an advantage. Washington has since restored limited Mythos access to some allies, but interest in sovereign AI capacity has continued to rise.

South Korea’s plan is to adapt a locally developed frontier model by adding security-related information to its training corpus. Bae said the resulting security-capable model should debut by the end of the year.

The government is also seeking bids for two other projects:

Recommended reading

Hassabis says STEM makes you 10x better at AI

• a free chatbot for all residents
• an agentic application to help people interact with government services

Bae made the remarks during a policy briefing led by President Lee Jae Myung. AI discussions there also covered using the technology to detect fake news in real time and to handle complaints about government services faster than current systems allow.