South Korea plans security-focused AI model by end of 2026

South Korea is building a domestic AI model aimed at cybersecurity work, with a target launch by the end of 2026, Deputy Prime Minister and Minister of Science and ICT Bae Kyung-hoon said. The project matters because Seoul is trying to reduce dependence on U.S.-controlled advanced models for bug finding, after access to Anthropic’s Mythos was restricted by Washington.

Bae disclosed the plan during a policy briefing led by President Lee Jae Myung. He said South Korea wants a model in the same class as Mythos, Anthropic’s security-oriented system, so the country has its own capability for finding software vulnerabilities. The government did not disclose the budget, compute partners, model architecture, dataset size or whether private AI labs will have access to the finished system.

The South Korean approach, according to Bae, is to add security-related material to the training corpus for a locally developed frontier model. That makes the effort less a standalone cybersecurity product launch and more an attempt to turn a national foundation-model program into a security asset.

U.S. access controls are the backdrop

The push follows two U.S. government interventions affecting Mythos access, according to Bae’s account. In one case, Washington required Anthropic to make the model available only to U.S. citizens, a condition the company could not satisfy, leading it to block access broadly. In another, the U.S. ordered Anthropic to suspend services while officials examined claims about possible dangerous performance issues.

Washington has since permitted limited Mythos access for some allies. Even so, those episodes have changed how other governments view reliance on U.S.-based AI systems for security work. If the U.S. can restrict access to a model with national-security value, domestic agencies and companies in countries without their own equivalents may be left behind when advanced capability is needed.

The Register reported that it is aware of another effort to create Mythos-like tools involving private companies and infrastructure operators across multiple countries. That suggests South Korea’s program is part of a broader move toward sovereign AI for security tasks, although the contours of those parallel efforts remain unclear.

More public-sector AI projects are planned

Bae also said South Korea has sought bids for a chatbot that would be free for all residents. The government is also looking for an agentic application to help people interact with public services. Details on vendors, procurement value and deployment dates were not disclosed.

AI was a broader theme of the presidential briefing. Officials discussed using the technology to detect fake news in real time and to respond to complaints about government services faster than current processes allow. Those are broad public-sector AI ambitions, and the government did not provide performance baselines or implementation details.

For technology suppliers, the practical signal is procurement demand around sovereign models, security datasets and government-facing AI agents. For policymakers, South Korea’s plan reflects a harder view of model access: strategic AI capabilities may be available to allies until they are not, and Seoul wants its own option before that becomes an operational problem.

This story draws on original reporting from The Register.