The participation of the South Korean government and major companies like Samsung Electronics and SK Hynix in Anthropic's AI cybersecurity initiative, Project Glasswing, marks a new phase in domestic infrastructure security collaboration. With the integration of semiconductor, telecommunications, and public security networks into a global AI security framework, cooperation in vulnerability detection and patching using AI is expected to expand.
According to the IT industry on June 3, Anthropic announced on June 2 (local time) that it would expand the number of institutions eligible to access the Claude Mythos preview under Project Glasswing to approximately 150 new organizations across 15 countries.
The new participants include key infrastructure sectors such as financial services, cybersecurity, and technology. Industries that were not initially included, such as power, water, telecommunications, healthcare, and hardware, have also been added, including the South Korean government and companies.
The Ministry of Science and ICT is participating through the Korea Internet & Security Agency (KISA), and it is reported that Samsung Electronics, SK Hynix, and SK Telecom (SKT) are also involved.
Project Glasswing is a closed security consortium created by Anthropic to identify and improve security vulnerabilities using the next-generation AI model, Mythos Preview. Mythos Preview is a high-performance AI model capable of understanding and analyzing complex software to detect vulnerabilities.
In April, Anthropic announced the Mythos Preview, providing access only to select global tech giants like Google, Microsoft, Apple, and Amazon Web Services (AWS) due to security concerns.
In response, governments of major countries, including South Korea, have raised concerns about the AI threat response capabilities related to Mythos. On May 11, South Korea's Deputy Minister of Science and ICT, Ryu Je-myung, met with Anthropic's Global Policy Head, Michael Salito, to explore the possibility of joining Project Glasswing.
With the inclusion of the South Korean government and major companies in the Glasswing initiative, there is optimism that this will enhance the country's capacity to respond to AI security threats, such as cyber vulnerability detection. Samsung Electronics and SK Hynix are key players in the global AI semiconductor supply chain, and both participated as "strategic infrastructure partners" in the Series H funding round that Anthropic secured on May 28 (local time).
The inclusion of KISA, a government agency, increases the likelihood of establishing a collaborative AI security network between private key companies and public security response systems.
Industry experts believe that Anthropic's expansion of participating institutions and companies is related to the findings of the first report from Project Glasswing. According to the initial results released on May 22 (local time), over 10,000 security vulnerabilities classified as "High" or "Critical" were identified in major software systems worldwide, which could be immediately exposed to hacking attacks.
In response, KISA issued a notice on May 25 through the Protect Korea and KrCERT websites, requesting enhanced security checks and readiness in light of concerns about the expansion of AI-based cyber threats. The agency provided guidelines for companies on how to prepare for AI-based cyber threats and recommended actions for CEOs.
KISA stated, "According to Anthropic's Project Glasswing investigation, over 6,200 high-risk vulnerabilities have been found in open-source software, indicating a growing threat from AI-based security issues," and urged companies to strengthen their security monitoring systems, including external anomaly traffic detection and blocking.
Previously, the Ministry of Science and ICT also formalized participation in OpenAI's trust-based access program for governments and institutions (GTAC). With the discussions surrounding Glasswing, South Korea's AI-based cybersecurity collaboration is expanding to include partnerships with global AI companies like OpenAI and Anthropic.
The security industry believes that as AI models evolve, the speed of vulnerability discovery and attack execution is increasing, making traditional preemptive security systems insufficient. Therefore, there is a need to establish systems for vulnerability detection, threat analysis, patch verification, and incident response.
Anthropic emphasized the necessity of preparing countermeasures, stating, "Within 6 to 12 months, other AI companies will likely have models comparable to Mythos, and these models may be released without adequate safety measures."
* This article has been translated by AI.
Copyright ⓒ Aju Press All rights reserved.