[Edaily Reporter Kim Hyun-ah ] The competition among leading domestic AI companies in the field of security AI is heating up. While a consortium comprising Naver (NAVER(035420)) and LG Corp. has been awarded the government’s security-specialized AI project, SKTelecom(017670)and Upstage have formed a separate working group to develop their own security AI.

While Naver and LG Corp. are pursuing a strategy of specializing their 700B-class AI in attack and defense, respectively, SKT and Upstage are focusing on “practical AI” designed to perform on-site tasks such as actual security monitoring and vulnerability analysis.

As concerns about AI-powered cyberattacks grow—following recent evidence that the autonomous attack tool ARTEX was used in hacking incidents targeting the financial sector—the focus of the security AI competition is shifting from model scale to actual attack response capabilities.

On the 8th, SKTelecom launched the “Independent AI Cybersecurity Consortium” in partnership with Upstage and six security firms—SK Shields, AHNLAB, SecuI, RaonSecure Co., Ltd., GENIANS, INC., and Piolink, Inc.—as well as three universities: Korea University, Soongsil University, and Pusan National University. A total of 11 organizations are participating.

SKT and Upstage Take a Separate Approach Amid Naver-LG Corp. Contract Win SKT

and Upstage competed against the Naver-LG Corp. consortium for the development of security AI under the government’s “Specialized AI Foundation Model Development” project. After Naver and LG Corp. were awarded the contract, SKT and Upstage launched their own security AI development initiative through a separate consortium.

SKT’s next-generation model, “A.X K3,” and Upstage’s next-generation “Sola” will be trained on real-world data from the field, including malware and vulnerability analysis as well as security monitoring. The plan is to develop AI that goes beyond simple security Q&A to identify vulnerabilities, apply patches, and respond to security incidents.

In particular, the project will utilize real-world data from domestic security firms, including SK Shields’ 17 billion daily threat events and AHNLAB,INC.’s more than 2.5 PB of security data.

The developed models will be integrated into the participating companies’ security products and used for tasks such as zero-day vulnerability analysis and real-time monitoring. Performance will be enhanced by incorporating real-world experience back into the training process.

Naver and LG Corp.: “Attack AI vs. Defense AI”

Since winning the government contract, Naver and LG Corp. have been developing two 700B-class security AI systems based on Naver’s HyperClova X and LG Corp.’s ExaOne, respectively.

Naver is strengthening its defense capabilities by utilizing 4,000 GPUs, while LG Corp. is focusing on developing offensive AI that simulates actual hacking attacks. The two models are connected via a “harness,” creating a structure where one attacks while the other defends.

Meanwhile, SKT and Upstage are introducing “mutual red teaming,” in which A.X K and Sola attack each other. They enhance defensive capabilities by using discovered vulnerabilities as training data.

During the development of A.X K2, SKT generated and trained on approximately 15,000 attack response data points through its proprietary red-teaming system, SERT. AHNLAB,INC., a participating company, possesses over 2.5 PB (petabytes) of security data, as well as malware and endpoint threat detection technologies. Korea University, Soongsil University, and Pusan National University independently evaluate the models’ security and performance from a third-party perspective.

The Decisive Factor Is ‘Real-World Attack Response Capability’

Naver and LG Corp. have secured approximately 830 TB of source data through collaboration with 21 organizations, including KEPCO KDN, Korea Hydro & Nuclear Power, and the Korea Financial Security Institute, and plan to apply the model to major strategic industries and critical national infrastructure.

SKT and Upstage are pooling real-time threat intelligence and operational experience with security solutions held by security firms, and plan to expand the number of participating organizations in the future.

While both camps utilize large-scale data, their strategies differ. While Naver and LG Corp. are focused on creating large-scale security models equipped with both offensive and defensive capabilities, SKT and Upstage are placing greater emphasis on verifying performance by deploying their models in real-world security environments.

The decisive factor in the cybersecurity AI competition is expected to be not the size of the model, but how quickly it can detect actual attacks, identify vulnerabilities, and respond to incidents.

Yoo Kyung-sang, Head of SK Telecom’s AI CIC, said, “We will demonstrate the effectiveness of models validated through our proprietary AI foundation model project in real-world security environments, in collaboration with leading domestic information security companies and universities, thereby showcasing the success of national AI models in the cybersecurity domain as well.”