Theori Korea is joining the Cybersecurity-Specialized AI Foundation Model Development project led by Korea's Ministry of Science and ICT (MSIT).

MSIT launched the initiative to counter cyber threats growing more sophisticated alongside advances in high-performance AI, and to establish independent Korean capability in security-specialized AI technology. Following a competitive review, a consortium led by NAVER Cloud was selected. As a core member of that consortium, Theori Korea will take on central R&D work alongside leading Korean security and ICT companies, universities, and research institutes.

Theori Korea brings to the project the vulnerability research and analysis experience and real-world data it has accumulated in offensive security, along with the agent-based vulnerability detection and validation technology it has developed through building and operating Xint, its autonomous AI security platform.

Theori Korea's specific mandate covers the harness that enables AI models to use real security tools and environments to detect, reproduce, and verify vulnerabilities, as well as the SFT (Supervised Fine-Tuning) and RLVR (Reinforcement Learning with Verifiable Rewards) environments used to train and advance the model on real-world hacking scenarios. The aim is a model that goes beyond generating security knowledge — one that solves problems in live environments and learns from results that can be independently verified.

The government will provide the consortium with 256 B200 GPUs over 10 months beginning this month. NAVER Cloud's consortium intends to use that capacity to build a security-specialized foundation model with a 700B-class Mixture-of-Experts (MoE) architecture.

"What matters in cybersecurity AI isn't just how much security knowledge a model holds, but whether it can actually find and verify vulnerabilities in a real environment," said Brian Pak, CEO of Theori. "Building on the hands-on agents and vulnerability validation technology we've developed through Xint, we want to help Korean AI models reach world-class capability in solving cybersecurity problems."