Why is Anthropic's 'Mythos' triggering an alert that has even the U.S. administration on edge?

Anthropic, a U.S. artificial intelligence (AI) startup, has sparked debate over AI and cybersecurity threats with its cutting-edge AI model 'Mythos' (Mythos). Proponents argue that powerful AI models can help identify security vulnerabilities in public- and private-sector systems in advance, while others worry that they could instead leave systems more exposed to AI-enabled cyberattacks. The U.S. administration, up to and including the vice president, is moving urgently, seeking responses to AI risks not only at the government level but also together with major Big Tech and financial institutions.

As of the 12th, according to industry sources, Mythos is an AI model that Anthropic has released only to about 40 to 50 companies on a limited basis to prevent potential misuse by hackers, stating that "its ability to detect software security vulnerabilities exceeded that of most people except top experts." Through 'Project Glasswing', launched with more than ten technology, security, and financial companies including Nvidia, Amazon, Cisco, and JPMorgan Chase, Anthropic agreed to jointly conduct "defensive (cyber) security operations" based on a preview (preview) version of Mythos.

Some suggest that Anthropic is exaggerating Mythos's power as a matter of corporate promotion. However, Mythos, which has shown high performance across various metrics such as coding and reasoning, has indeed achieved groundbreaking results in detecting software defects (bugs). It demonstrated the ability to uncover large numbers of 'zero-day vulnerabilities'which even developers or system administrators have not recognized and therefore have no security patchessuch as discovering a bug in the OpenBSD operating system for the first time in 27 years. It is also reportedly capable of automatically generating exploit code after detecting a bug. In other words, one cannot rule out the possibility that non-experts without hacking skills could use Mythos to infiltrate systems.

In response, the U.S. administration held countermeasure meetings at various levels after Mythos was released. The Wall Street Journal (WSJ) reported on the 11th (local time) that "the risks of AI have emerged as a top priority for the Donald Trump administration," and that White House National Cyber Director Sean Kean-Cross is working with officials from relevant departments to identify security vulnerabilities in critical national infrastructure and to strengthen government systems that AI could exploit.

Earlier, Vice President JD Vance and Treasury Secretary Scott Bessent also held a conference call with heads of Big Tech and major banks to discuss response measures to potential AI-driven cyberattacks. Federal Reserve (Fed) Chair Jerome Powell also took part in the call.

Mantas Mazeika, a researcher at the AI Safety Institute, told the Christian Science Monitor that "a comprehensive recognition of the cyber risks posed by AI has begun."

Reuters-Yonhap News

한글기사 원본(Original Korean Story)