KT and SK Telecom Lead Korean AI Sovereignty With Open Source Language Models
KT launches Trust 2.0, an open-source Korean AI model supporting data sovereignty and cultural nuance. Partnering with Microsoft, KT balances local innovation and global collaboration.

KT Advances Korean AI with Open-Source 'Trust 2.0' and Strategic MS Partnership
KT is making significant strides in Korean artificial intelligence (AI) development by launching its super-large language model (LLM) named Trust 2.0. This model, tailored for Korean language and culture, will be released as open source on Hugging Face, fostering accessibility and collaboration among developers, startups, and institutions.
By adopting open source, KT encourages improvements, customizations, and wider use without restrictions. This approach aims to reduce development costs and accelerate innovation within Korea’s AI ecosystem.
Strategic Two-Track AI Development
KT’s strategy involves building its own AI model while maintaining collaboration with global tech leaders like Microsoft. This dual approach supports Korea’s push for Sovereign AI, a concept emphasized by the current government to secure AI technologies developed domestically for national security and economic independence.
Shin Dong-hoon, KT’s Chief AI Officer, highlights four guiding principles for AI development at KT:
- Data sovereignty within Korea
- User choice with diverse AI options
- Embedding Korean values in AI services
- Responsible and regulation-compliant AI operation
Technical Strengths of Trust 2.0
Trust 2.0 builds on the foundation of Trust 1.0, which has been used in KT’s AI Customer Center, Genie TV content recommendations, and AI phone guidance. The upgraded model is trained on high-quality, carefully curated Korean data, classified into over 200 categories with strict filtering to remove unreliable sources.
The model is available in mini and base versions, with plans for a professional version. It supports both Korean and English, utilizing KT’s proprietary Talk Nizer — a tool that breaks down sentences into tokens optimized for AI processing.
On the Ko-Sovereign benchmark, Trust 2.0 outperformed comparable domestic and international models, particularly excelling in tests focused on Korean language and cultural nuances. It can analyze conversational context deeply, such as interpreting emotions and implied meanings beyond simple summarization.
Sovereign AI: Securing Korea’s AI Future
Sovereign AI refers to AI systems developed and controlled domestically, essential for sectors like defense, healthcare, pharmaceuticals, manufacturing, and content creation. KT stresses that without Sovereign AI, Korean IT services risk losing leverage and becoming overly dependent on global tech giants.
KT acknowledges collaboration with Microsoft for technologies it hasn’t yet developed internally but emphasizes this partnership is to achieve optimal AI performance without compromising its Sovereign AI goals.
Comparison: SK Telecom’s AOTX 4.0
On the same day, SK Telecom released its open-source model AOTX 4.0, which focuses on enhanced Korean language processing and on-premise deployment for security. AOTX 4.0 scored higher than OpenAI’s ChatGPT-4 on Korean benchmarks such as KMMLU2 and CLIcK3, proving its effectiveness in Korean contexts.
SK Telecom developed AOTX 4.0 entirely with internal data training, securing data sovereignty without external dependencies. It offers two model sizes: a standard 72 billion parameter version and a lighter 7 billion parameter model, giving users flexibility based on their needs.
What This Means for IT and Development Professionals
KT’s Trust 2.0 and SK Telecom’s AOTX 4.0 models reflect a growing emphasis on AI models tailored specifically for the Korean language and culture, with open-source availability accelerating innovation. For IT professionals, these developments offer opportunities to engage with state-of-the-art Korean AI models that prioritize sovereignty, data security, and localized performance.
Leveraging open-source AI models like Trust 2.0 can reduce development time and costs while aligning with national standards for data handling and AI ethics. The coexistence of proprietary collaboration and open development environments creates a balanced ecosystem for Korean AI progress.
For more insights on AI development and practical training, explore resources at Complete AI Training.