How Will Anthropic’s Claude Mythos Change Cybersecurity in Asia?

Translation result.

Anthropic’s next‑generation artificial intelligence (AI) model, Mitos, has upended the global cybersecurity landscape, and South Korea’s government is exploring ways to obtain related intelligence through a U.S.‑led big‑tech cooperative. Officials say the move would buy critical time to respond to the security threats posed by ultra‑capable AI models such as Mitos.

According to security industry and government sources on April 19, the Ministry of Science and ICT (MSIT) is seeking ways to gain proactive access to Mitos‑related data collected by Project Glasswing. Mitos is the high‑end model Anthropic unveiled on April 7. Anthropic warned that Mitos’s capabilities are so advanced that a public release could be weaponized by hackers, so it offered early access to a limited set of companies and organizations to develop mitigation strategies. That consortium is Project Glasswing. About 50 participants, including 12 major tech firms such as Amazon Web Services (AWS), Microsoft, Google, Apple, Cisco and Palo Alto Networks, are using the preview build.

MSIT launched the review after a sharp rise in concern among domestic institutions and firms following Mitos’s debut. Anthropic says Mitos found a vulnerability in OpenBSD—a system long regarded as among the most secure—that had gone undetected for 27 years. The model can autonomously detect vulnerabilities and even generate and execute exploit code. “Experts have long warned that advances in AI would dramatically increase both the volume and speed of cyberattacks, and this is that warning realized,” said Kim Jin‑soo, president of the Korea Information Security Industry Association. “Mitos is likely only the first such model; similar systems will emerge from other companies and states.” On April 14, OpenAI also said it would give select researchers early access to a software‑security‑vulnerability model dubbed ‘GPT‑5.4‑Cyber.’

The industry problem is twofold: tools now exist that allow non‑expert users to breach government and financial networks, and access to those tools is concentrated among U.S. companies. Glasswing participants can run Mitos against their systems, discover vulnerabilities and roll out patches quickly. Non‑participating firms and states, by contrast, must prepare on their own until Anthropic publishes the report it says it will release in July.

At an emergency meeting MSIT recently convened with academics and industry leaders, participants urged the government to step in. The Financial Services Commission also told MSIT it wants access to Glasswing findings. “It’s unrealistic for individual Korean firms to join Glasswing one‑by‑one,” one expert at the meeting said. “If the government can use diplomatic channels to obtain mitigation know‑how—even informally—that would be beneficial.” An MSIT official confirmed the ministry is considering access options but said no specific plan or decision on information sharing has been made. Politico reports that, aside from the U.K., few European governments have tested Mitos at the institutional level.

Experts say the government should pursue short‑term countermeasures for Mitos while also building long‑term security governance for an era of AI‑enabled hacking. Yoon Du‑sik, CEO of Irewon & Company, said, “Getting Mitos‑related information could speed defensive responses by months, but that’s only a stopgap. At the national level, authorities should stand up a control tower to marshal resources for security response. Companies and agencies must inventory all IT assets and rethink how they use on‑premises software and internal networks.”