Korea Seeks Access to Glasswing to Counter Mithos Hacking Risks | #hacking | #cybersecurity | #infosec | #comptia | #pentest | #hacker

As Anthropic’s next-generation artificial intelligence (AI) model “Claude Mithos” shakes up the global cybersecurity landscape, the Korean government is pursuing ways to receive related information through a cooperation body centered on U.S. Big Tech firms. The move aims to buy time for responding to security threats that ultra-high-performance AI models such as Mithos could bring.

According to the security industry and the government on Wednesday, the Ministry of Science and ICT (MSIT) is exploring ways to preemptively access Mithos-related information collected by “Project Glasswing.” Mithos is the top-tier AI model unveiled on the 7th of this month by Anthropic, the developer of Claude. Anthropic said Mithos is so powerful that releasing it to the public carries the risk of being exploited for hacking, adding that it would first allow certain companies and organizations to use it while working out response strategies. That consortium is Project Glasswing. About 50 companies and organizations, including 12 Big Tech firms such as Amazon Web Services (AWS), Microsoft, Google, Apple, Cisco, and Palo Alto Networks, are participating and using the preview version.

MSIT launched the review because anxiety has surged among domestic institutions and companies following the emergence of Mithos. According to Anthropic, Mithos discovered a vulnerability in OpenBSD — regarded as one of the most secure operating systems — that had gone undetected for 27 years. It also autonomously performs tasks ranging from detecting vulnerabilities to writing and executing attack code. Kim Jin-soo, chairman of the Korea Information Security Industry Association, said, “The outlook that AI advances would rapidly increase the volume and speed of cyberattacks has long been raised, but it has now materialized.” He added, “Mithos is just the beginning. Similar models will emerge from other companies and countries.” Indeed, OpenAI also announced on the 14th of this month that it would first offer a “GPT-5.4-Cyber” model specialized in detecting software (SW) security vulnerabilities to a select group of experts.