This page was machine-translated and may differ from the original. View original
SKT and Upstage Launch AI Cybersecurity Council...11 Organizations Participate

Advancing Proprietary AI Model with 6 Security Companies and 3 Universities
SK Telecom and Upstage are joining hands with major domestic information security companies and universities to strengthen AI-based cybersecurity capabilities. The plan is to train proprietary AI models with actual security field data and establish a virtuous cycle of applying them to participating companies' security products.
SK Telecom and Upstage announced on the 8th that they have launched the 'Proprietary AI Cybersecurity Council' comprising a total of 11 organizations, including 6 information security companies such as SK Shieldus, Secui, AhnLab, Raon Secure, Genians, and Piolink, as well as 3 universities including Korea University, Soongsil University, and Pusan National University, and have concluded a memorandum of understanding.
The signing ceremony was held on the 7th at the SKT Tower.
This council was formed to expand SKT's A.X K series and Upstage's Solar, which have advanced to phase 3 of the government's 'Proprietary AI Foundation Model Project (Dokpamo)', into the cybersecurity domain.
The council operates by training two AI models with security field data held by participating organizations to enhance their capability in performing security tasks.
Training proceeds in two stages: pre-training and post-training.
In pre-training, the model's foundational knowledge is strengthened with security data, and SKT's next-generation model 'A.X K3' will be reinforced with data for safe operation of the 'AI for All' service along with security domain data.
Upstage focuses on training vulnerability detection and analysis data on Solar's next-generation model, which is capable of large-scale code and long-term log analysis.
Post-training enhances practical capabilities such as △software vulnerability analysis and patching △security operations center incident response △safe use of security tools.
The scale of field data from participating companies is also substantial.
SK Shieldus analyzes 17 billion threat events in real-time daily through its integrated security management platform 'Secudium'.
AhnLab possesses over 2.5PB (petabytes) of security data and malware and endpoint threat detection technologies.
Secui provides high-performance firewall solutions, Raon Secure provides autonomous penetration testing technology, Genians provides post-training datasets, and Piolink provides autonomous attack validation and detection rule generation technology, respectively.
The council applies 'mutual red-teaming' in which A.X K and Solar attack and evaluate each other to identify vulnerabilities.
During A.X K2 development, SKT built its own red-teaming automation system SERT (Self-Expanding Red-Teaming) and generated and trained approximately 15,000 attack response datasets, which will be extended to mutual red-teaming.
Upstage utilizes 210,000 high-difficulty and safety data and alignment methodologies to strengthen security functions.
Participating universities are responsible for red-teaming design and evaluation from a third-party perspective.
Korea University validates the AI's executable file analysis capability and defensive strength, Soongsil University validates security performance evaluation criteria and responses to guardrail bypass attacks, and Pusan National University validates the performance and safety of next-generation and lightweight models, respectively.
Yoo Kyung-sang, SKT AI CIC Head, said, "Models verified in the proprietary AI foundation model project will demonstrate effectiveness in actual security fields together with Korea's representative information security companies and universities, showcasing the utilization achievements of the nation's representative AI model in the cybersecurity domain as well."
To request a correction, reply or follow-up report on this article, see how to file a request. Previously published statements are collected in corrections & replies.












